Author: Frank Phillips
Date: 15:29:53 02/16/05
Go up one level in this thread
On February 16, 2005 at 16:26:50, Robert Hyatt wrote: >On February 16, 2005 at 15:57:46, Andrew Williams wrote: > >>On February 16, 2005 at 15:33:24, Robert Hyatt wrote: >> >>>On February 16, 2005 at 14:28:57, Dieter Buerssner wrote: >>> >>>>On February 16, 2005 at 13:19:46, Robert Hyatt wrote: >>>> >>>>>On February 16, 2005 at 13:09:14, Frank Phillips wrote: >>>> >>>>>>I have been using the profile generated optimisation option, but the code it >>>>>>produces is no faster then with simple -O3. >>>> >>>>I have no experience on AMD64. On x86 it works well for me, under Linux and >>>>Windows. >>>> >>>>>ICC (Intel's compiler) works fine and that is what I use >>>>>myself. >>>> >>>>Does profile guided optimizatation with ICC work for you, when you run crafty >>>>under xboard. It doesn't work for me. >>> >>>Never tried. I do my profile runs via a "Make profile" where I have a target >>>that compiles for profiling, runs a bunch of test cases, then compiles using the >>>data gathered to improve the code. So my profiling is all in "command mode" as >>>well, which is just as good since the xboard stuff is not in the search of my >>>program at all anyway. >>> >>> >>>> Console runs work fine. With gcc, I can >>>>rund a match Yace-Crafty under Xboard. That gave better speedups than >>>>test-suites. But no luck with ICC here. I get very modest speedups with PGO and >>>>ICC (more with gcc). However, ICC is fast even without PGO. >>> >>>I can't get gcc to compile and run crafty with profiling optimizations. It will >>>produce the profile data files fine, but when I go to re-compile with the >>>options to use the profile data, it complains about one or more of the profile >>>data files being corrupted. I've tested this hundreds of times since this was >>>put into gcc, no luck at all. I seem to recall I might have gotten it to work >>>on the opteron last year, but I don't believe there was any speedup if I did, >>>and I might well be remembering that wrong anyway. >>> >>>But gcc won't profile-optimize for me period. although it will profile just fine >>>and produce output that helps in analyzing performance. >>> >>> >>> >> >>I had this phenomenon and I got rid of it by removing an optimization option. >>I'm afraid I don't remember which one, but I think it was -fomit_frame_pointer; >>I discovered what the problem was while reading a website, which I can no longer >>find. >> >>Andrew > > >I think I did this. It seems to me that -fomit-frame-pointer produces errors >when used with any sort of profiling, but I am not sure... > >I'll try again to see... Try just -fprofile-arcs / -fprofile-brach-probabilities instead of generate/use, which if I seem to remember did not work in at least one version of gcc3.3 (Mandrake 10.0 IIRC). Frank
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.