Testing result of run-time after loop-parallelization

It's unlucky that blogger is unavailable  now in China that I could only
using gladder to get access. But it's lucky that blogger has a email
post functionality, so, it's fine.

I have tested some simple loops which parallelized by autopar
to see how much promotion we can got. This test results now
is available in gcc wiki:
http://gcc.gnu.org/wiki/AutoparRelated.
Mainly include:

1. One nested loop, number of threads vs. run-time, with
   not all the code parallel.
2. One nested loop, number of threads vs. run-time, almost
   all code got parallel.
3. Two nested loops, with innermost parallel, number of threads
    vs. run-time. (It got worse after parallel.)
4. Two nested loops, with outer loop parallel, number of threads
    vs. run-time. (It's better).

The 4 figures include comparing of:
 - Runtime of openmp directive inserted c code.
 - Runtime of Graphite autoparallel c code.
 - Runtime of non-parallel c code.

For this week, I'm planning to do some work with
the already-done data dependency work. Maybe firstly write
some test cases for the data dependency checking to see what
kinds of loops we could/want parallelize.