Yet another tree question

Yet another tree question

[David Dolan]

Hello,

First, I would like to thank everyone who has helped me thus far, I really
appreciate it.

Now here comes the good part :)

Basically, the situation I'm in is this: I'm working on a research project
doing data dependency analysis.  We want to use GCC to help us with this
task.  This is the first time I have ever done anything related to
compilers, so I'm learning as I go.  Now, what we really want to do is use
the GCC tree structure to do our analysis on (basically, take the tree
before it is put into RTL form and work with that).  My job is to determine
the structure of the tree (and also if it is actually what we need).

Now, the problem is, I am only a summer intern.  Since the summer is waning
away, so is my job.  This leaves me with not enough time to wade through
1,000's of lines of code to find what I need.  Which brings me to my
question...Tim Josling provided me with his documentation on interfacing the
tree structure.  Thanks Tim, but I guess what I'm really looking for is
something more along the lines of "a declaration node is of the form..."
documentation.

I'm sorry to be such a slacker, and if you don't want to help me on that
principle alone, that's fine, I can understand that.  But, if anyone knows
if something like this exists, you would be a life-saver :)  Conversely, if
anyone knows that nothing like this exists, please let me know.

Thanks for enduring my email!!  And for your help!!!!

Dave Dolan
ddolan@andrew.cmu.edu

Re: Yet another tree question

[Tim Josling]

All there is that I could find, is the doco links on the gcc
site, Joachim Nadler's and my tree documentation and the source
code. I would also like to know if there is anything else. 

As for whether the tree structure is suitable for data analysis,
it is a fairly low level representation and this may be a
hindrance. Also The structures tend to disappear quickly (eg at
end of function) so doing whole program analysis is problemantic.
Also the tree structures are very 'C' oriented so if you are
analysing anything other than C/++ it will be less useful.

The files tree.h and tree.def are pretty well commented so they
are well worth a look.

Tim Josling


David Dolan wrote:

> Dave Dolan
> ddolan@andrew.cmu.edu

Re: Yet another tree question

[Benedetto Proietti]

Hi David,
...

David Dolan wrote:
> 
> Hello,
> 
> First, I would like to thank everyone who has helped me thus far, I really
> appreciate it.
> 
> Now here comes the good part :)
> 
> Basically, the situation I'm in is this: I'm working on a research project
> doing data dependency analysis.  We want to use GCC to help us with this
> task.  This is the first time I have ever done anything related to
> compilers, so I'm learning as I go.  Now, what we really want to do is use
> the GCC tree structure to do our analysis on (basically, take the tree
> before it is put into RTL form and work with that).  My job is to determine
> the structure of the tree (and also if it is actually what we need).

Don't know anything more what people already told you!

 
> Now, the problem is, I am only a summer intern.  Since the summer is waning
> away, so is my job.  This leaves me with not enough time to wade through
> 1,000's of lines of code to find what I need.  Which brings me to my
> question...Tim Josling provided me with his documentation on interfacing the
> tree structure.  Thanks Tim, but I guess what I'm really looking for is
> something more along the lines of "a declaration node is of the form..."
> documentation.

Try to take a look to the Sage++ project:
http://www.extreme.indiana.edu/sage/

It is a source code manipulator: it parses C++ code into a tree
representation,
with classes to navigate through it.
You can add or modify code, ie. to unroll loop or to do your data
dependency analysis; then you unparse the tree back into C++ code that
you can give to GCC.

The project *seems* to be not manteined but the is some documentation.
We are starting using it to encapsulate language extensions for a
parallel machine so to limit extensions to gcc, and also, of course, for
optimizations.


> I'm sorry to be such a slacker, and if you don't want to help me on that
> principle alone, that's fine, I can understand that.  But, if anyone knows
> if something like this exists, you would be a life-saver :)  Conversely, if
> anyone knows that nothing like this exists, please let me know.
> 
> Thanks for enduring my email!!  And for your help!!!!

> Dave Dolan
> ddolan@andrew.cmu.edu


Come on, keep trying !!
...and good luck!

Benedetto Proietti
INFN - APE
benedetto.proietti@roma1.infn.it