This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [new-regalloc] Dataflow too slow...
- To: Michael Hayes <m dot hayes at elec dot canterbury dot ac dot nz>
- Subject: Re: [new-regalloc] Dataflow too slow...
- From: Daniel Berlin <dberlin at redhat dot com>
- Date: Sat, 3 Feb 2001 18:13:54 -0500 (EST)
- cc: Geert Bosch <bosch at gnat dot com>, <gcc at gcc dot gnu dot org>
On Sun, 4 Feb 2001, Michael Hayes wrote:
> Geert Bosch writes:
> > One of the things I'm stumbling on now is that df_rd_global_compute
> > takes way too much time to allow me to bootstrap the Ada compiler.
>
> I used a simple worklist algorithm for this to get the ball rolling.
> The usual solution to impove the performance is to order the blocks
> with a depth first search. This was on my todo list.
I did this as well, i forgot to mention.
I use flow_compute_depth_first_order to get the dfs ordering, and then
have it pick the first block on the worklist, in order of dfs order
IE
for (i=0; i < n_basic_blocks; i++)
if (TEST_BIT (worklist, dfs_order[i])
return dfs_order[i];
>
> There are many other more esoteric algorithms in the literature which
> would be a nice self-contained project for someone to implement.
Yup.