branches compared (lja_speed, EV56)

[Kurt Garloff <garloff at suse dot de>]

Hi,

lja_speed benchmarks on my EV56-600 (DEC21164A), 768MB, KDE kicker 
background load, times in seconds.

                        -fno-new-ra  -fnew-ra	-fnew-ra 
						-fbranch-prob
*** With FFM
gcc-3.2.2 (SL81)          14.06		 9.37+	  9.24+
gcc-3.2.3                 14.12  	 
gcc-334     20040409      13.84		 9.46+	  9.25+ 
gcc-333 ham 20040409      16.56-	13.81-	 13.72 
gcc-34      20040409      12.89		11.18 
34 new_ra   20040409      16.67-	14.14-   ERROR! 
gcc-35      20040409      13.03		11.25	 ERROR!
35 tree-ssa 20040416	  12.01+	11.81	 14.10-
35 ssa-lno  20040409	  13.08		12.64
*** With CPML
gcc-3.2.2 (SL81)          14.18 	 9.27+		
ccc-6.5.9                  8.60

Compile flags: 
gcc: -Wall -O2 -ffast-math -fomit-frame-pointer -fschedule-insns2 -O3 \
 -frerun-loop-opt -mcpu=ev56 -funroll-loops -fstrict-aliasing 
ccc: -w0 -msg_display_tag -O2 -accept restrict_keyword \
 -D__USE_STD_IOSTREAM -fast -tune ev56 -arch ev56 -O4 -inline speed 

Notes:
* With the old register allocator, you can see improvements
  from 3.2 -> 3.3 -> 3.4 -> ssa.
  Both 3.5 and ssa-lno are a bit behind, hammer branch and new_ra
  branches suck.
* With the new register allocator, 3.3.4 and 3.2.2SUSE perform best,
  3.4, 3.5 are acceptable, ssa somewhat worse, ssa-lno worse
  and hammer and new-ra branches suck.
* The hammer branch slowdown is probably an import from 3.4; it used
  to be absent 9 month ago, then a slowdown appeared on 3.4, which is
  meanwhile cured.
* FDO is broken on 3.4 and 3.5.

Regards,
-- 
Kurt Garloff  <garloff@suse.de>                            Cologne, DE 
SUSE LINUX AG, Nuernberg, DE                          SUSE Labs (Head)

Attachment: lja_speed.c
Description: Text document

Attachment: pgp00000.pgp
Description: PGP signature