[graphite] fix loop domain matrix construction

[Sebastian Pop]

Hi,

I committed the attached fix for avoiding to include in the loop
domain matrix loops that are not around the innermost considered loop.
So now instead of building constraint matrices with zeros for columns
representing loops that are not around, i.e. not in the domain of the
considered loop, like this:

  eq   l1   l2  cst
   1   -1    0   98
   1    1    0    0

in this example l2 is not in the domain of loop l1, we generate:

  eq   l1  cst
   1   -1   98
   1    1    0

that produces a nicer code without an infinite loop for l2 ;-)

for (s_0=0;s_0<=98;s_0++) {
  S1 ;
}

This is something that Tobias asked me to fix some time ago.  Thanks
for the bug report ;-)

Sebastian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 916_graphite.diff
Type: text/x-diff
Size: 10826 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20080206/567969c0/attachment.bin>