This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Pessimization from reduce-all-givs and move-all-movables
- From: Jerry Quinn <jlquinn at us dot ibm dot com>
- To: gcc at gcc dot gnu dot org
- Date: Wed, 20 Nov 2002 01:07:20 -0500
- Subject: Pessimization from reduce-all-givs and move-all-movables
I'm trying to see how much speed I can squeeze out of a 1.8GHz
Pentium4 laptop. I've been playing with the c4 benchmark
(http://www.cwl.nl/~tromp/c4/fhour.html) and have been trying out
various optimization combos.
One of the combos I tried was:
gcc-dev -fmove-all-movables -freduce-all-givs -O2 -mcpu=pentium4
-march=pentium4 -msse2 -DUNIX -DTRANSIZE=1050011 -DPROBES=8
-DREPORTPLY=8 *.c
which gives a score of 976 Kpos/sec.
Without -fmove-all-movables -freduce-all-givs, performance improves to
1048 Kpos/sec, a 7% difference.
The docs ask for examples of code that is pessimized by these two
options, so here is some.
Also, -O3 loses about 1% on this code. And a 700MHz Pentium3 laptop
can be tuned to a score of 812 with the mainline. While the best I've
been able to get out of this machine is 1100. There are a bunch of
shifts, so that may be the main source of slowness.
Jerry Quinn