llvm dev - Jan 2016 - Can someone give me some pointer on alias analysis ?

2015-12-26 18:32 GMT+01:00 Philip Reames <listmail at philipreames.com>:
> On 12/26/2015 02:17 AM, Amaury SECHET via llvm-dev wrote:
>
> I'm trying to fix that bug: https://llvm.org/bugs/show_bug.cgi?id=20049
>
> It turns out this is the kind of optimization that I really need, as when
> it isn't done, all kind of other optimizations opportunities down the
road
> are not realized as they are not exposed.
>
> I have no idea where to start digging for this. I assume there is some
> kind of interaction between memory dependency and alias analysis that is
> overly conservative. Can someone gives me some infos on how the 2 interact
> together ? What is the code that is supposed to remove these kind of loads
?
>
> First, it looks like you're using an older version of LLVM (the load
> syntax has changed).  That's definitely not recommended since it will
> greatly limit your choices when encountering an issue like this.
>
>
I just reused an old code sample. The problem still exists in recent
version of LLVM.

> Second, the aliasing problem has to do with the effects of the
> "allocmemory" callsite on a memory location associated with an
earlier
> alloca.  There's nothing that prevents your allocmemory function from
> writing to global state.  You have to teach the alias analysis that an
> unescaped noalias pointer can't alias the global state allocmemory
might
> access.  Slightly surprised we don't get this today, but oh well.  Take
a
> look at the isNonEscapingLocalObject predicate in BasicAA.  Then look at
> "getModRefInfo(ImmutableCallSite CS, const MemoryLocation
&Loc)".  Double
> check to make sure this is the one that MDA actually calls.
>
>
I don't think this is the problem. When there is only 2 calls to
allocmemory, loads are optimized away as expected. But it seems that the
analysis is confused with 3+ calls.

> Third, you might want to take a look the new inaccessiblememonly
> attribute. Depending on how you're modelling your allocation, this
might
> help resolve the aliasing problem in different way.  However, be aware that
> this is *very* new.  In particular, I triggered an optimizer bug by adding
> the new attribute to your particular example.  :(
>
> Fourth, have you considered implementing a simple escape analysis pass?  I
> notice that the store-load forwarding would just fall out once you removed
> the last allocation.  I believe the fixed point would then become a
> function which returns the constant answer and does nothing else.
>
>
Yes I did. More specifically, there is a pass that try to recognize memory
allocation like calls and optimize them, which right now have libc and some
overloads of operator new hardcoded in it but not much more. This could be
greatly improved IMO, to be language agnostic (and better support C++ in
the process), but really, I do think relying on this to get the load
eliminated, when noalias already provide this information to the optimizer,
is overkill.

> Fifth, some pointers on debugging this yourself.  GVN (which is the one
> which does complicated *-load forwarding) calls MDA
> (MemoryDepedenceAnalysis).  Using the appropriate -debug-only options will
> likely get you to the right area.  You can also consider using the
> MemDepPrinter pass to bypass GVN.  I don't know of a way to issue raw
AA
> queries for testing.  That would be useful, but I don't know that we
have
> it.
>
> Hope that helps.
>
>
Yes, thanks. I'll probably come back with more question once I've played
with these options a bit. Sorry for the late answer, I was mostly off the
grid the past 2 weeks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160104/91d3ad68/attachment.html>

After a bit more investigation, it turns out that because %0 is stored into
%1 (after bitcast) and so %3 may have access to it and clobber it.

After a bit of thought, it is correct in the general case, but definitively
something stricter is needed here. Looking at inaccessiblememonly I'm not
sure this is what is needed. What if the memory allocator is defined is the
current module ?

This leads me to conclude this is way more linked to the memory allocation
pass than I expected it to be in the first place. Can I ask what you plan
to use inaccessiblememonly for ? Should the semantic be refined to fit the
bill better ?

2016-01-04 12:29 GMT+01:00 Amaury SECHET <deadalnix at gmail.com>:
>
>
> 2015-12-26 18:32 GMT+01:00 Philip Reames <listmail at
philipreames.com>:
>
>> On 12/26/2015 02:17 AM, Amaury SECHET via llvm-dev wrote:
>>
>> I'm trying to fix that bug:
https://llvm.org/bugs/show_bug.cgi?id=20049
>>
>> It turns out this is the kind of optimization that I really need, as
when
>> it isn't done, all kind of other optimizations opportunities down
the road
>> are not realized as they are not exposed.
>>
>> I have no idea where to start digging for this. I assume there is some
>> kind of interaction between memory dependency and alias analysis that
is
>> overly conservative. Can someone gives me some infos on how the 2
interact
>> together ? What is the code that is supposed to remove these kind of
loads ?
>>
>> First, it looks like you're using an older version of LLVM (the
load
>> syntax has changed).  That's definitely not recommended since it
will
>> greatly limit your choices when encountering an issue like this.
>>
>>
> I just reused an old code sample. The problem still exists in recent
> version of LLVM.
>
>
>> Second, the aliasing problem has to do with the effects of the
>> "allocmemory" callsite on a memory location associated with
an earlier
>> alloca.  There's nothing that prevents your allocmemory function
from
>> writing to global state.  You have to teach the alias analysis that an
>> unescaped noalias pointer can't alias the global state allocmemory
might
>> access.  Slightly surprised we don't get this today, but oh well. 
Take a
>> look at the isNonEscapingLocalObject predicate in BasicAA.  Then look
at
>> "getModRefInfo(ImmutableCallSite CS, const MemoryLocation
&Loc)".  Double
>> check to make sure this is the one that MDA actually calls.
>>
>>
> I don't think this is the problem. When there is only 2 calls to
> allocmemory, loads are optimized away as expected. But it seems that the
> analysis is confused with 3+ calls.
>
>
>> Third, you might want to take a look the new inaccessiblememonly
>> attribute. Depending on how you're modelling your allocation, this
might
>> help resolve the aliasing problem in different way.  However, be aware
that
>> this is *very* new.  In particular, I triggered an optimizer bug by
adding
>> the new attribute to your particular example.  :(
>>
>> Fourth, have you considered implementing a simple escape analysis pass?
>> I notice that the store-load forwarding would just fall out once you
>> removed the last allocation.  I believe the fixed point would then
become a
>> function which returns the constant answer and does nothing else.
>>
>>
> Yes I did. More specifically, there is a pass that try to recognize memory
> allocation like calls and optimize them, which right now have libc and some
> overloads of operator new hardcoded in it but not much more. This could be
> greatly improved IMO, to be language agnostic (and better support C++ in
> the process), but really, I do think relying on this to get the load
> eliminated, when noalias already provide this information to the optimizer,
> is overkill.
>
>
>> Fifth, some pointers on debugging this yourself.  GVN (which is the one
>> which does complicated *-load forwarding) calls MDA
>> (MemoryDepedenceAnalysis).  Using the appropriate -debug-only options
will
>> likely get you to the right area.  You can also consider using the
>> MemDepPrinter pass to bypass GVN.  I don't know of a way to issue
raw AA
>> queries for testing.  That would be useful, but I don't know that
we have
>> it.
>>
>> Hope that helps.
>>
>>
> Yes, thanks. I'll probably come back with more question once I've
played
> with these options a bit. Sorry for the late answer, I was mostly off the
> grid the past 2 weeks.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160104/c2eea738/attachment.html>

On 01/04/2016 03:29 AM, Amaury SECHET wrote:
>
>
> 2015-12-26 18:32 GMT+01:00 Philip Reames <listmail at philipreames.com 
> <mailto:listmail at philipreames.com>>:
>
>     On 12/26/2015 02:17 AM, Amaury SECHET via llvm-dev wrote:
>>     I'm trying to fix that bug:
>>     https://llvm.org/bugs/show_bug.cgi?id=20049
>>
>>     It turns out this is the kind of optimization that I really need,
>>     as when it isn't done, all kind of other optimizations
>>     opportunities down the road are not realized as they are not
exposed.
>>
>>     I have no idea where to start digging for this. I assume there is
>>     some kind of interaction between memory dependency and alias
>>     analysis that is overly conservative. Can someone gives me some
>>     infos on how the 2 interact together ? What is the code that is
>>     supposed to remove these kind of loads ?
>     First, it looks like you're using an older version of LLVM (the
>     load syntax has changed).  That's definitely not recommended since
>     it will greatly limit your choices when encountering an issue like
>     this.
>
>
> I just reused an old code sample. The problem still exists in recent 
> version of LLVM.
>
>     Second, the aliasing problem has to do with the effects of the
>     "allocmemory" callsite on a memory location associated with
an
>     earlier alloca.  There's nothing that prevents your allocmemory
>     function from writing to global state. You have to teach the alias
>     analysis that an unescaped noalias pointer can't alias the global
>     state allocmemory might access.  Slightly surprised we don't get
>     this today, but oh well.  Take a look at the
>     isNonEscapingLocalObject predicate in BasicAA.  Then look at
>     "getModRefInfo(ImmutableCallSite CS, const MemoryLocation
&Loc)".
>     Double check to make sure this is the one that MDA actually calls.
>
>
> I don't think this is the problem. When there is only 2 calls to 
> allocmemory, loads are optimized away as expected. But it seems that 
> the analysis is confused with 3+ calls.
>
>     Third, you might want to take a look the new
>     inaccessiblememonlyattribute. Depending on how you're modelling
>     your allocation, this might help resolve the aliasing problem in
>     different way.  However, be aware that this is *very* new.  In
>     particular, I triggered an optimizer bug by adding the new
>     attribute to your particular example.  :(
>
>     Fourth, have you considered implementing a simple escape analysis
>     pass?  I notice that the store-load forwarding would just fall out
>     once you removed the last allocation.  I believe the fixed point
>     would then become a function which returns the constant answer and
>     does nothing else.
>
>
> Yes I did. More specifically, there is a pass that try to recognize 
> memory allocation like calls and optimize them, which right now have 
> libc and some overloads of operator new hardcoded in it but not much 
> more. This could be greatly improved IMO, to be language agnostic (and 
> better support C++ in the process), but really, I do think relying on 
> this to get the load eliminated, when noalias already provide this 
> information to the optimizer, is overkill.
It turns out we already have something along these lines in 
InstCombine.  Take a look at visitAllocSite in 
InstructionCombining.cpp.  It's rough and good use some refinement, but 
its a good starting place for a generalized escape
analysis.
>
>     Fifth, some pointers on debugging this yourself.  GVN (which is
>     the one which does complicated *-load forwarding) calls MDA
>     (MemoryDepedenceAnalysis).  Using the appropriate -debug-only
>     options will likely get you to the right area.  You can also
>     consider using the MemDepPrinter pass to bypass GVN.  I don't know
>     of a way to issue raw AA queries for testing.  That would be
>     useful, but I don't know that we have it.
>
>     Hope that helps.
>
>
> Yes, thanks. I'll probably come back with more question once I've 
> played with these options a bit. Sorry for the late answer, I was 
> mostly off the grid the past 2 weeks.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160104/d3459458/attachment.html>

On 01/04/2016 07:32 AM, Amaury SECHET wrote:
> After a bit more investigation, it turns out that because %0 is stored 
> into %1 (after bitcast) and so %3 may have access to it and clobber it.
Can you give a bit more context?  I'm not sure which of the examples 
you're talking about.
>
> After a bit of thought, it is correct in the general case, but 
> definitively something stricter is needed here. Looking at 
> inaccessiblememonly I'm not sure this is what is needed. What if the 
> memory allocator is defined is the current module ?
At the moment, inaccessiblememonly would require separate compilation of 
the allocation function.
>
> This leads me to conclude this is way more linked to the memory 
> allocation pass than I expected it to be in the first place. Can I ask 
> what you plan to use inaccessiblememonly for ? Should the semantic be 
> refined to fit the bill better ?
Well, I didn't introduce the attribute, so I can't speak for the 
original intent.  For me, I plan on applying it to some of our out of 
line allocation functions and other helper routines which modify runtime 
state, but not java visible state.

If you have specific suggestions for how to refine the semantics, please 
make them.  Getting the details right is always the hard part.  :)

You might also consider using a variant of your allocation function 
which takes a pointer to the global state it needs to modify.  Doing 
this would allow you to use argmemonly to restrict the aliasing while 
still allowing whole program optimization.  I haven't tried this in 
practice, but it seems like it would probably work...
>
> 2016-01-04 12:29 GMT+01:00 Amaury SECHET <deadalnix at gmail.com 
> <mailto:deadalnix at gmail.com>>:
>
>
>
>     2015-12-26 18:32 GMT+01:00 Philip Reames
>     <listmail at philipreames.com <mailto:listmail at
philipreames.com>>:
>
>         On 12/26/2015 02:17 AM, Amaury SECHET via llvm-dev wrote:
>>         I'm trying to fix that bug:
>>         https://llvm.org/bugs/show_bug.cgi?id=20049
>>
>>         It turns out this is the kind of optimization that I really
>>         need, as when it isn't done, all kind of other
optimizations
>>         opportunities down the road are not realized as they are not
>>         exposed.
>>
>>         I have no idea where to start digging for this. I assume
>>         there is some kind of interaction between memory dependency
>>         and alias analysis that is overly conservative. Can someone
>>         gives me some infos on how the 2 interact together ? What is
>>         the code that is supposed to remove these kind of loads ?
>         First, it looks like you're using an older version of LLVM
>         (the load syntax has changed).  That's definitely not
>         recommended since it will greatly limit your choices when
>         encountering an issue like this.
>
>
>     I just reused an old code sample. The problem still exists in
>     recent version of LLVM.
>
>         Second, the aliasing problem has to do with the effects of the
>         "allocmemory" callsite on a memory location associated
with an
>         earlier alloca.  There's nothing that prevents your
>         allocmemory function from writing to global state.  You have
>         to teach the alias analysis that an unescaped noalias pointer
>         can't alias the global state allocmemory might access. 
>         Slightly surprised we don't get this today, but oh well.  Take
>         a look at the isNonEscapingLocalObject predicate in BasicAA. 
>         Then look at "getModRefInfo(ImmutableCallSite CS, const
>         MemoryLocation &Loc)".  Double check to make sure this is
the
>         one that MDA actually calls.
>
>
>     I don't think this is the problem. When there is only 2 calls to
>     allocmemory, loads are optimized away as expected. But it seems
>     that the analysis is confused with 3+ calls.
>
>         Third, you might want to take a look the new
>         inaccessiblememonlyattribute. Depending on how you're
>         modelling your allocation, this might help resolve the
>         aliasing problem in different way.  However, be aware that
>         this is *very* new.  In particular, I triggered an optimizer
>         bug by adding the new attribute to your particular example.  :(
>
>         Fourth, have you considered implementing a simple escape
>         analysis pass?  I notice that the store-load forwarding would
>         just fall out once you removed the last allocation.  I believe
>         the fixed point would then become a function which returns the
>         constant answer and does nothing else.
>
>
>     Yes I did. More specifically, there is a pass that try to
>     recognize memory allocation like calls and optimize them, which
>     right now have libc and some overloads of operator new hardcoded
>     in it but not much more. This could be greatly improved IMO, to be
>     language agnostic (and better support C++ in the process), but
>     really, I do think relying on this to get the load eliminated,
>     when noalias already provide this information to the optimizer, is
>     overkill.
>
>         Fifth, some pointers on debugging this yourself. GVN (which is
>         the one which does complicated *-load forwarding) calls MDA
>         (MemoryDepedenceAnalysis).  Using the appropriate -debug-only
>         options will likely get you to the right area.  You can also
>         consider using the MemDepPrinter pass to bypass GVN.  I don't
>         know of a way to issue raw AA queries for testing.  That would
>         be useful, but I don't know that we have it.
>
>         Hope that helps.
>
>
>     Yes, thanks. I'll probably come back with more question once
I've
>     played with these options a bit. Sorry for the late answer, I was
>     mostly off the grid the past 2 weeks.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160104/d04247f9/attachment.html>