This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Fix PR23744, VRP not merging discontinuous ranges of PHIs


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Duncan Sands wrote on 05/16/06 06:30:

> You are correct, this may be a problem.  I will try to do some benchmarking,
> but it may take a few days.  In the meantime, Richard Guenther has kindly
> agreed run the patch through his C++ tester.
> 
Luckily, I was wrong.

Compile times are mostly in the noise (times in seconds):

		Before	After	% change
cc1-i-files	594.46	597.34	+0.5%
DLV		214.00	215.17	+0.5%
SPEC2000	403.25	408.85	+1.4%
TRAMP3D		 92.87	 96.36	+3.8%


The effects of the new meet operator are negligible but mostly positive,
so I think we should keep it.  But let's wait for the next stage 1.

Predicates folded
- -----------------

cc1-i-files	13,217	13,257	+0.30%
DLV		 3,148	 3,148
SPEC2000	 1,554	 1,583	+1.87%
TRAMP3D		   834	   835	+0.12%


Constants propagated
- --------------------

cc1-i-files	33,519	33,547	+0.08%
DLV		 4,864	 4,864
SPEC2000	 9,980	10,006	+0.26%
TRAMP3D		13,192	13,191	-0.01%


Copies propagated
- -----------------

cc1-i-files	10,080	 9,972	-1.07%
DLV		 4,681	 4,677	-0.09%
SPEC2000	13,457	13,280	-1.32%
TRAMP3D		 3,036	 3,035	-0.03%
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)

iD8DBQFEaevkUTa2oAUaiwQRApDnAJ9Bt0cKt8oAv76n1Z+i57PXqy3oQgCdEk+3
ikSnjEESls/0nOnCT7dwGfo=
=DIdk
-----END PGP SIGNATURE-----


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]