Author: Robert Hyatt
Date: 11:18:39 08/27/05
Go up one level in this thread
On August 27, 2005 at 04:40:33, Andreas Guettinger wrote: >On August 26, 2005 at 20:32:16, Robert Hyatt wrote: > >>On August 26, 2005 at 18:11:39, Andreas Guettinger wrote: >> >>>On August 26, 2005 at 15:32:08, Robert Hyatt wrote: >>> >>>>On August 26, 2005 at 15:26:27, Andreas Guettinger wrote: >>>> >>>>>On August 26, 2005 at 15:09:29, Robert Hyatt wrote: >>>>> >>>>>>I just released version 20.0. The only new feature beyond 19.20, is that the >>>>>>book files (*.bin) are now portable. That is, you can build a book on a >>>>>>big-endian architecture, and use it on a little-endian architecture. You can >>>>>>build a book on a 32 bit architecture, and use it on a 64 bit architecture. >>>>>>Etc. This was a sore point with me for quite a while... some systems/compilers >>>>>>make structures multiples of 8 bytes, some multiples of 4, some are big-endian, >>>>>>some are little-endian. In the 2004 WCCC blitz tournament, I gave peter a book >>>>>>that was made on my 32 bit systems, which produced some bizarre stuff on the 64 >>>>>>bit opteron. I decided to fix that. >>>>>> >>>>>>20.1 will be out as soon as I get all the bit renumbering done. Nothing new >>>>>>other than a very slight expected speedup due to the removal of the 63 - bit# >>>>>>translation required between the Intel BSF/BSR and the Cray LEADZ bit numbering >>>>>>schemes... >>>>>> >>>>>>20.2 and beyond are a planned complete revision of the Evaluation, but these >>>>>>changes will not be released publicly until after the next WCCC event... >>>>> >>>>>Does this mean we should download a new book.bin file too? >>>>> >>>>>regards >>>>>Andy >>>> >>>> >>>>No. The "standard format" is the x86 book.bin, which is what everyone has >>>>always downloaded from my web site. If you are running on anything other than a >>>>32 bit X86 architecture, you had to build a book.bin for that architecture. You >>>>now need to either download the "standard format" or else build a new one, which >>>>will be in standard format automatically. >>> >>>Strange. If I remember correctly I downloaded the book.bin file from your ftp >>>some time ago and used it all the time with crafty19.19 on my amd64 system. Are >>>you sure its not the x86_64 version? >> >> >>Were you running a 64 bit operating system? And a version of crafty compiled >>for 64 bits? That's where the rub comes in. And, in fact, the book will appear >>to "sorta work" but most of the possible moves are not available. > >Oh. I didn't think about that. I thought it would crash or not recognize the >book at all. But it "sorta worked" at least. :) > >Yes, I'm running on a 64bit linux system and compiling with linux-amd64. As I >wrote in my other answer, I built my own book now with crafty-20.0, which was >not possible for me on previous versions, because of a segmentation fault at the >very end of the merging process. Ran into that on AMD as well. Turns out there was a very long name in the enormous.pgn file that was overflowing the pgn tag buffer. Not sure if it was white or what. Had to debug it prior to the WCCC in fact, and found it. Made the PGN tag buffers longer to solve the problem... > >Something else. What are these .gcda files that can't be found during compiling? >Seems to be triggered by -fbranch-probabilities. > >Andy Yes. If you are using gcc 3.3.3, you can do a "make linux-profile" and then run some test positions, then touch *.c *.cpp and do a "make linux" this uses profile-guided optimizations to improve the execution speed. I also have a "make profile" target, but it uses the wrong make options for you. If you look, right at the top of the "profile:" target, replace the make linux-profile-icc with "make linux-profile, and farther down, replace the make linux-icc with make linux. It takes about 5 mins to compile, profile and re-compile...
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.