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: [RFT] fwprop merge


	I have run SPEC CPU2000 on a PowerMac G5 with and without the
fwprop patch.

BASE is -O2 -mcpu=G5 -mdynamic-no-pic -malign-natural
PEAK is -O3 -mcpu=G5 -ffast-math -funroll-loops -fpeel-loops -mdynamic-no-pic -malign-natural

The percent difference in the SPEC score is:
		BASE	PEAK
gzip:		0%	1%
vpr:		0%	0%
gcc:		0%	0%
mcf:		0%	0%
crafty:		0%	0%
parser:		0%	0%
eon:		0%	3%
perlbmk:	-1%	-1%
gap:		-9%	-3%
vortex:		0%	0%
bzip2:		0%	0%
twolf:		0%	0%
wupwise:	0%	0%
swim:		0%	0%
mgrid:		0%	0%
applu:		-1	1%
mesa:		-2%	-2%
galgel:		0%	0%
art:		0%	0%
equake:		0%	0%
facerec:	0%	0%
ammp:		0%	0%
lucas:		0%	0%
fma3d:		0%	0%
sixtrack:	0%	0%
apsi:		0%	0%

	Other than gap, the effect of the patch is in the noise.  The
decrease in performance can be addressed after the patch is merged.

	I also am rerunning the SPEC tests without the 32/64 mixed mode.

David


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