This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: BCT optimization
- To: Michael Hayes <m dot hayes at elec dot canterbury dot ac dot nz>
- Subject: Re: BCT optimization
- From: David Edelsohn <dje at watson dot ibm dot com>
- Date: Sun, 20 Sep 1998 15:12:27 -0400
- Cc: egcs at cygnus dot com, Jeffrey Law <law at cygnus dot com>
One of the reasons for the regressions on some loop cases in the
testsuite when using the BCT optimization is that insert_bct() does not
appear to do as good a job at calculating the number of iterations as
unroll.c:loop_iterations() for the constant int case. It also is not
being as careful about the signedness of loop bounds. I am not sure if
the runtime calculation of loop iterations is correct, but insert_bct()
definitely misses some compile time calculations.
I am going to convert the BCT optimization to use
loop_iterations() when it can. I also am fixing the formatting in many
places which did not follow the GNU coding standards.
If anybody else plans to make changes to the BCT optimization
functions in loop.c in the near future, please coordinate that with me so
we don't trample each other and make merging mokre difficult than it needs
to be.
Thanks, David