On Jun 6, 9:32 am, "Dr. David Kirkby" <[EMAIL PROTECTED]> wrote: > On Jun 6, 7:44 am, mabshoff <[EMAIL PROTECTED]> wrote:
Hi David, > > > The Solaris port does seem to be coming on, which is good. The ntl fix > > > from Francois seems to work on both my SPARCs. I do have another > > > problem, but that needs to be on another subject, not this one. > > > Thanks for testing ;) > > I should have added the problem I got was with 3.0.2, with twisted. I > think you have perhaps made some changes to that, so I'm in the > process of downloading sage-3.0.3.alpha1.tar. Unfortunately, I'm only > getting 50 KB/s, it will be another 15 minutes before I have the > tarball. That tarball issue with twisted has not been fixed yet since I have not dropped the recompressed tar ball into alpha1 yet. The box I am doing that is behind a firewall that blew up yesterday, but will hopefully be operational again today. > > But the medium term plan is to switch to the Sun Forte compiler and I > > have no intention to ever even put substantial time to make the 64 bit > > port work with gcc. I have had to deal with too much crap with gcc in > > 64 bit mode on Solaris to even pursuit that, but patches and fixes are > > obviously welcome. Since the Sun compiler is free as in beer I don't > > think that is too much of a problem. Should Sun decide to do something > > stupid and force people to pay for their compilers again we will > > revisit gcc in 64 bit mode on Solaris again. > > Essentially the Sun compilers are faster than gcc, so changing for > that reason alone would be useful on code which is going to be CPU > intensive. Well, I have seen those benchmark numbers but I will remain skeptical until I ran Sage benchmarks myself. I once bought into the Intel C++ compiler creating 40% faster code than the gcc compiler [two, three years back it was claimed that mysql compiled with Intel's C++ compiler was that much faster compated with the then current gcc release] and for the code I was working with it turned out that gcc was on average 2% faster than Intel's C++ compiler with CPU specific tuning options and all bells & whistles like feedback optimization turned on. > I can't see Sun trying to charge for their compilers again. I think > they would **** off too many people, including many of their own > employees. But who knows, they have made some odd decisions before > with how they changed the license conditions on Solaris. Yes, Sun has done a lot of odd things in the last decade and their somewhat uneasy hot and cold relationship with Linux has certainly not made them too many friends in certain OS circles. Charging for their development tools must be one of the dumbest moves Sun ever did. It made people use gcc and so much software no longer compiled with the Sun compiler any more. Since gcc could not produce 64 bit code on Solaris for a long time it also didn't make Sparc more attractive either. Sparc's raw CPU power sucks anyway as is, but fortunately Sun now figured out that making x86-64 boxen is good business ;) > * At one time Solaris was commerical only > > * Then it became free for up to 8 CPUs. It did not inforce the 8-CPU > limit, but few people have machines with more than 8 CPUs. > > * Then they called Solaris free, but in only if the machine was > capable of holding just one cpu, and furthermore the machine was > bought from a Sun authorised resller. This mean it was not free for > 99% of users. In practice, since anyone could download it, you can > guess what people did. > > * At one point they used to charge a nominal fee to download the x86, > but not SPARC version. I think they got fed up with a lot of > downloads, with people not bothering to install it. I can see some > logic in that. > > * Then Solaris became free and open source. Well, you still have to register for the privilege to download security fixes and that is annoying. I am sure that many people once they get Solaris up and running will be turned off by that. Solaris 10's package selection mechanism is a joke [unfulfilled dependencies are not resolved automatically on request, but you can try until you can figure it out :)], but Indiana will hopefully fix that. > So who knows? Depending on what way the wind is blowing, Sun will send > you a Solaris Express DVD and them pick up the postage, or you pay > for the postage. Their policy seems to change from week to week, with > no real logic to it. > > Going back to Sage, I imagine the switch from gcc->Sun compilers is > going to present quite a few challenges, as the options are so > different. I believe there is a gcc interface to the Sun compilers so > it can be called with gcc options, but I think it is better to use the > right Forte options. I don't see much trouble in fixing the various flags, but then I have worked with the Sun compilers for a while. > BTW, are there any other ports planned, like HP-UX? I have a HP-UX box > here, but it has not been switched on for many years. I also have > tru64 on a Dec Alpha, IRIX on an SGI Octane and AIX on a large IBM > server, but none of the machines have been on for years. About time I > got rid of them I guess! I have access to a multi CPU Dec Alpha with loads of RAM, but I don't see any benefit from porting besides the potential bugs exposed. For SageNet I have asked for HPUX/Itanium, PPC 64/AIX and PPC 64/ Linux. I am not sure if or when any of those will show up. Somebody (Oskar) is currently porting Sage to IRIX for fun. In the end * Windows x86, x86-64 * OSX x86, x86-64, ppc, ppc 64 * Solaris x86, x86-64, Sparc * Linux x86, x86-64, Itanium will cover the vast majority of available hardware out there. > But I know from my efforts on ATLChttp://atlc.sourceforge.net/that building > software on multiple > platforms tends to find a lot of bugs, which don't show up on platform > A, but the bugs are there, ready to surface at some later date. ATLC > has been built on many platforms including a Sony Playstation and a > Cray supercomputer. I agree with you that this is a big benefit of porting to "miscellaneous" platforms and the port to Linux/ppc for example has turned up some interesting bugs that later popped up elsewhere (#1337 is a good example). That being said once we have the high priority port targets working [basically the list above] I am more than interested to port to something else provided you gave a shell account for me to play with. But we do cover quite a large range of hardware and finding targets that are viable is getting harder and harder. And it doesn't look like we are running out of bugs to fix anytime soon ;) Cheers, Michael --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
