top output (compilation took about 10 minutes)
# cc1 8.6% 5:13.31 1 13 365 730M+ 2.45M 519M- 1.13G
time to run all tests (3rd launch):
real 0m2.713s
user 0m1.321s
sys 0m0.306s
Without optimization flag (compilation takes 30 seconds):
# cc1 84.0% 0:17.15 1 13 203 243M+ 4.42M 245M+ 435M+
time to run all tests (3rd launch)
real 0m2.560s
user 0m1.323s
sys 0m0.319s
I think it would be a good change, especially when people compile this
on a server with limited memory and shared CPU resources...
PATCH:
Index: ext/superredcloth_scan/extconf.rb
==================================================================---
ext/superredcloth_scan/extconf.rb (revision 245)
+++ ext/superredcloth_scan/extconf.rb (working copy)
@@ -1,9 +1,6 @@
require ''mkmf''
-if /darwin9/ =~ RUBY_PLATFORM
- # OSX 10.5 doesn''t like the default "-Os"
- $CFLAGS << " -O1 "
-end
+$CFLAGS << " -O0 " # do not optimize (takes too much memory and
performance gain is negligeable)
dir_config("superredcloth_scan")
have_library("c", "main")
Gaspard: Hey, wow! You''re my new favorite person. I''d been commenting out nonessentials when I was working on something specific and just waiting the 10 minutes when I wanted to see the whole thing work. I''m committing this change. Johan, since you''re the one who requested the -O1 flag, would you verify it still works for you in Leopard? If anyone knows a reason it''s bad to compile with -O0, now''s the time to speak up. Great work, Gaspard! Thanks! Jason On Feb 27, 2008, at 1:32 AM, Gaspard Bucher wrote:> top output (compilation took about 10 minutes) > # cc1 8.6% 5:13.31 1 13 365 730M+ 2.45M > 519M- 1.13G > > time to run all tests (3rd launch): > real 0m2.713s > user 0m1.321s > sys 0m0.306s > > Without optimization flag (compilation takes 30 seconds): > # cc1 84.0% 0:17.15 1 13 203 243M+ 4.42M 245M+ > 435M+ > > time to run all tests (3rd launch) > real 0m2.560s > user 0m1.323s > sys 0m0.319s > > I think it would be a good change, especially when people compile this > on a server with limited memory and shared CPU resources... > > PATCH: > Index: ext/superredcloth_scan/extconf.rb > ==================================================================> --- ext/superredcloth_scan/extconf.rb (revision 245) > +++ ext/superredcloth_scan/extconf.rb (working copy) > @@ -1,9 +1,6 @@ > require ''mkmf'' > > -if /darwin9/ =~ RUBY_PLATFORM > - # OSX 10.5 doesn''t like the default "-Os" > - $CFLAGS << " -O1 " > -end > +$CFLAGS << " -O0 " # do not optimize (takes too much memory and > performance gain is negligeable) > > dir_config("superredcloth_scan") > have_library("c", "main") > _______________________________________________ > Redcloth-upwards mailing list > Redcloth-upwards at rubyforge.org > http://rubyforge.org/mailman/listinfo/redcloth-upwards
On Wed, Feb 27, 2008 at 12:22:54PM -0500, Jason Garber wrote:> I''m committing this change. Johan, since you''re the one who > requested the -O1 flag, would you verify it still works for you in > Leopard?I agree, nice tweak! I''m playing with the Ragel flags and noticing wildly different behaviors as well. (Using -G0, -T0, -T1, etc. instead of -G2.) I''d really like to assemble a rake tasks that would go through and try permutations of Ragel flags with GCC flags and determine the most optimal settings. For sure, though, compilation time was getting ridiculous. _why
While trying ragel optimization settings, it would be good to record the size of the output files, since I think it would be nice to include these files in the gem in order to avoid ragel dependency for those not modifying the gem. Gaspard 2008/2/28, why the lucky stiff <why at hobix.com>:> On Wed, Feb 27, 2008 at 12:22:54PM -0500, Jason Garber wrote: > > I''m committing this change. Johan, since you''re the one who > > requested the -O1 flag, would you verify it still works for you in > > Leopard? > > > I agree, nice tweak! I''m playing with the Ragel flags and noticing > wildly different behaviors as well. (Using -G0, -T0, -T1, etc. > instead of -G2.) I''d really like to assemble a rake tasks that > would go through and try permutations of Ragel flags with GCC flags > and determine the most optimal settings. > > For sure, though, compilation time was getting ridiculous. > > > _why > > _______________________________________________ > Redcloth-upwards mailing list > Redcloth-upwards at rubyforge.org > http://rubyforge.org/mailman/listinfo/redcloth-upwards >