Re: Re: Fwd: Re: More precise values in Analyzer

Eugene,

I looked at SourceValue and SourceInterpreter, and it appears that when a
SourceValue comes from a parameter (or this pointer or exception variable),
it just reports the source as an empty set.

Maybe some more background will help:

The goal of this project is to make an optimizing ahead-of-time compiler for
the Avian virtual machine (oss.readytalk.com).  I want to take in a list of
class files, and spit out lots of llvm assembly.  The input and output would
look roughly like this (as part of a larger program):

[javap style assembly]
public int returnFirstParameter(int i)
Code:
0:  iload 1
1:  ireturn

[llvm assembly]
define i32 @TestClasS_returnFirstParameter(i32 %i) nounwind {
entry:
  ret i32 %i;
}

LLVM (low-level virtual machine, llvm.org - essentially a compiler /
optimizer infrastructure) stores its IR completely in SSA form.  This
requires naming the parameters to a method, and tracking their uses.

Just using the variable slot number to detect a parameter may not be
sufficient, because a java program could (though they don't usually) decide
to overwrite its own arguments.  In this case, this overwritting value would
be detected as the original parameter, when in fact it is a new value.  I
need something that is correct 100% of the time, not just in the (very)
common cases.

Even if this would work, I don't see a way to get the specific local
variables index that a certain value originated in (particularly in the case
of Values allocated with Interpreter.newValue(...).  Am I missing something?

I've read significant portions of "ASM 3.0 A Java bytecode engineering
library <http://download.forge.objectweb.org/asm/asm-guide.pdf>", and I
didn't see anything that could help me with this particular problem.  Do you
have a certain section that you are thinking of?

Thanks,

Joshua



On Thu, Jul 15, 2010 at 3:51 PM, Eugene Kuleshov <[email protected]>wrote:

>
>  Maybe I don't understand your issue completely, but this and method param
> values are easy. Normally you can tell them by variable slot number.
>  Also, I was referring to the SourceInterpreter, which records exactly
> where values came from. Not the best javadoc there, but look at the users
> guide book for more details and examples.
>  I guess there is a bit of gap, i.e. we don't have an Interpreter that
> provides both the type and source at the same time.
>
>  regards,
>  Eugene
>
>
> Joshua Warner wrote:
>
>> I accidentally replied to you directly, instead of the mailing list...
>>
>> ---------- Forwarded message ----------
>> From: *Joshua Warner* <[email protected] <mailto:
>> [email protected]>>
>> Date: Thu, Jul 15, 2010 at 2:16 PM
>> Subject: Re: [asm] Re: More precise values in Analyzer
>> To: Eugene Kuleshov <[email protected] <mailto:[email protected]>>
>>
>>
>> Eugene,
>>
>> Perhaps I don't understand what you mean, but the only information passed
>> to the Interpreter.newValue method is the Type of the requested value.
>>
>> The problem I am having is more related to values that aren't produced by
>> opcodes / instructions - such as the parameters in the local variable slots
>> and the "this" pointer.  I need to be able to distinguish when a Value
>> generated through newValue() is a parameter, "this" pointer or uninitialized
>> value.
>>
>> Sincerely,
>>
>> Joshua
>>
>>
>> On Thu, Jul 15, 2010 at 2:05 PM, Eugene Kuleshov <[email protected]<mailto:
>> [email protected]>> wrote:
>>
>>    Joshua,
>>
>>     Isn't value source can be derived from the opcode type that
>>    produced it?
>>
>>     regards,
>>     Eugene
>>
>>
>>    On Thu, Jul 15, 2010 at 3:56 PM,  <[email protected]
>>    <mailto:[email protected]>> wrote:
>>    >
>>    > Hi,
>>    >
>>    > Im trying to use ASM in a Java-to-LLVM translator project.  In
>>    short, I'm using
>>    > the Analyzer and a custom version of Value to (attempt to)
>>    extract a SSA form
>>    > of the Java bytecode.
>>    >
>>    > The problem I am having is that some of the origins of the
>>    values are not
>>    > precise enough.  For instance, at present, it seems impossible
>>    to track where
>>    > some of the values originally came from - Analyzer.analyze just
>>    calls
>>    > Interpreter.newValue(...) for uninitialized values, parameters,
>>    the this
>>    > reference, etc.
>>    >
>>    > What do you recommend?
>>    >
>>    > Would I be able to get a patch through that gives the
>>    Interpreter room to
>>    > distinguish between these?  I propose adding
>>    Interpreter.newParameter(int
>>    > index, Type type), Interpreter.newReturnValue(Type type),
>>    > Interpreter.newThisReference(Type type), and
>>    Interpreter.newUninitialized().
>>    >
>>    > FYI, I am using the stable 3.3 version - but the nightly SVN
>>    seems to have the
>>    > same problem.
>>    >
>>    > Sincerely,
>>    >
>>    > Joshua Warner
>>    >
>>    >
>>    > --
>>    > You receive this message as a subscriber of the [email protected]
>>    <mailto:[email protected]> mailing list.
>>
>>    > To unsubscribe: mailto:[email protected]
>>    <mailto:[email protected]>
>>    > For general help: mailto:[email protected]
>>    <mailto:[email protected]>?subject=help
>>    > OW2 mailing lists service home page: http://www.ow2.org/wws
>>    >
>>    >
>>
>>
>>    --
>>    You receive this message as a subscriber of the [email protected]
>>    <mailto:[email protected]> mailing list.
>>
>>    To unsubscribe: mailto:[email protected]
>>    <mailto:[email protected]>
>>    For general help: mailto:[email protected]
>>    <mailto:[email protected]>?subject=help
>>    OW2 mailing lists service home page: http://www.ow2.org/wws
>>
>>
>>
>>
>
>
> --
> You receive this message as a subscriber of the [email protected] mailing list.
> To unsubscribe: mailto:[email protected]
> For general help: mailto:[email protected]?subject=help
> OW2 mailing lists service home page: http://www.ow2.org/wws
>
>

-- 
You receive this message as a subscriber of the [email protected] mailing list.
To unsubscribe: mailto:[email protected]
For general help: mailto:[email protected]?subject=help
OW2 mailing lists service home page: http://www.ow2.org/wws