Compilers - Soliciting for C programs or snippets to test alias analysis

Author

Soliciting for C programs or snippets to test alias analysis

neal

2005-12-08, 3:59 am

Dear folk,

I'm working on a standalone pointer/alias analyzer for ANSI C
(C99), and have a plan to release the analyzer as open source project.

To ensure the correctness of the analysis, identify the cause of
imprecision, and explore the method to improve precision, I need some
real uses of pointer in C. Even though I have searched C programs on
sourceforge.net, it's really difficult to figure out which one is
suitable for my purpose. I'm particularly looking for code with the
following criteria:

*For standalone C programs*:
1. There is no restriction on the coding style, but it has to be
completely compatible with ANSI C (C89 or C99)
2. It can be compiled with open source library, or if you can
provide binary lib image.
3. The size of program should be just few thousand lines of C
code or at most 10,000 lines.
4. It heavily uses dynamically allocated memory.
5. You believe that there are opportunities in optimization if
"precise" alias information is provided.
option 6. if you know some variables are never aliased and can
list those variables.

*For C snippets*:
1. There are some reasons you think finding the anti-alias
information is very useful.

I sincerely solicit for your help. If you know any C programs or
snippets which satisfy the above criteria, please send to me. It would
be very helpful, if you can write a short description about the
functionality and especially the uses of pointer in your programs.
After I receive your program and descriptions, I will compile a
contribution list, and post both code and description online.

Thanks for your kindly help in advance.
Sincerely,
Neal

Derek M. Jones

2005-12-08, 10:12 pm

Neal,

> To ensure the correctness of the analysis, identify the cause of
> imprecision, and explore the method to improve precision, I need some
> real uses of pointer in C. Even though I have searched C programs on
> sourceforge.net, it's really difficult to figure out which one is
> suitable for my purpose. I'm particularly looking for code with the
> following criteria:

I imagine you will also want to compare the performance of your
analysis against that of other researchers. Why not use the same
programs that they use (invariably some version of the SPEC
benchmarks).

I don't understand why you are only interested in stand alone pointer
analysis. The purpose of pointer analysis is to provide information
to other kinds of analysis (eg, optimizers, lint like tools) to help
them do a better job.

> 3. The size of program should be just few thousand lines of C
> code or at most 10,000 lines.

This is a general problem with pointer analysis. Many algorithms do
not scale to industrial size programs.

neal

2005-12-11, 7:19 pm

Hello Derek,

Actually, performance is not my goal at this moment. I have
successfully analyzed some programs in SPEC95. But the size of those
programs is too large, it's really difficult to verify the correctness
of the analysis result. It's also hard to verify if SPEC benchmarks
are good for benchmarking pointer anlaysis. I believe we should have
some benchmarks especially designed for pointer analysis.

Best regards,
Neal

2005-12-11, 7:19 pm

Some other suites which partcularly use dynamic memory are - Ptrdist,
Olden, and mediabench. All of them are available online.

Standalone analyze I would argue has a lot of benefits,
especially researchers who work on compiler infrastructures which do
not have satisfactory alias analysis can use that. Most freely
available compiler infrastructures do not have satisfactory alias
analysis. Standalone will fill this gap....Do let the group know when
you have one, I would be keen to take a look ...

dz

Sponsored Links