This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Brad's code
- To: gcc at gcc dot gnu dot org
- Subject: Brad's code
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Sat, 09 Oct 1999 22:01:49 -0600
- Reply-To: law at cygnus dot com
A correction.
I couldn't find the message from rth which indicated the connectivity in
Brad's testcase that was compiling so slow.
I just found it ;-)
The cfg has 4895 basic blocks and 2264436 edges. So on average each node has
an out degree of 463. In a typical program the out degree of a node should be
less than 2. To quote rth "yowza".
[ I had used an out degree of 20 in my earlier message based on my (incorrect)
memory of the number of edges in the cfg. ]
jeff
ps. FWIW I'm planning to tweak the gcse code to not optimize functions with
absurdly high connectivity in the cfg. Solving dataflow equations in these
kinds of cases is going to take a very long time and provide marginal (if
any) benefit.