G
Greg Buchholz
Since gcc-4.1 now has the option to perform profile guided
optimization, I thought I'd try compiling a version of Ruby with this
feature turned on. For the (somewhat contrived/synthetic) scenario
below, it looks like compiling with profile guided optimization turned
on, along with -O3, results in about a 9% speed increase vs. -O3 alone.
YMMV. I compiled a version of Ruby-1.8.4 using the following config
options...
configure CFLAGS="-O3 -fprofile-generate"
LDFLAGS=-fprofile-generate
....Then I ran the short version of the n-body program from the Great
Computer Language Shootout with n=1000...
http://shootout.alioth.debian.org/debian/benchmark.php?test=nbody&lang=ruby&id=0
....this generated the profile information needed for the next stage. I
then compiled a Ruby with...
configure CFLAGS="-O3 -fprofile-use"
....Running the resulting interpreter runs the exact same n-body program
about 8-10% faster than an compiling with just -O3.
FWIW,
Greg Buchholz
optimization, I thought I'd try compiling a version of Ruby with this
feature turned on. For the (somewhat contrived/synthetic) scenario
below, it looks like compiling with profile guided optimization turned
on, along with -O3, results in about a 9% speed increase vs. -O3 alone.
YMMV. I compiled a version of Ruby-1.8.4 using the following config
options...
configure CFLAGS="-O3 -fprofile-generate"
LDFLAGS=-fprofile-generate
....Then I ran the short version of the n-body program from the Great
Computer Language Shootout with n=1000...
http://shootout.alioth.debian.org/debian/benchmark.php?test=nbody&lang=ruby&id=0
....this generated the profile information needed for the next stage. I
then compiled a Ruby with...
configure CFLAGS="-O3 -fprofile-use"
....Running the resulting interpreter runs the exact same n-body program
about 8-10% faster than an compiling with just -O3.
FWIW,
Greg Buchholz