[graphite] first part of the polyhedral representation

[Sebastian Pop]

Hi,

Here is the patch, that I will commit to the graphite branch.
The patch reworks most of the code that was in graphite.[ch].
The detection of the scops remains the same, and now we
detect parameters of the scop and build in function of these
parameters the iteration domains.  The access matrices are
not yet built, but shouldn't be difficult now.  I thought that I
have written some code for that, but I'm seeing that the
toplevel is disabled just now.  I'll rework on that part.

The notion of parameter for a scop is something
for which I didn't get a good definition.
Informally, a scop param is a declaration that
cannot be instantiated and that is defined before
the scop, ie. in a bb that precedes the entry of the
scop.  Now for avoiding to have too many
parameters I had to change the instantiate_parameters
function to walk further than stopping at the first variable
outside some loop.  I didn't liked the way I implemented
that part, and this will surely change, as I would like to
create a new instantiate function, for not disturbing the
other passes by this new behavior.

Sebastian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 640_graphite.diff
Type: text/x-patch
Size: 39279 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20061207/9d2b3b49/attachment.bin>