Re: RFC: Policy for arch specs
On Mon, 2 Jun 1997, Thomas Koenig wrote: > Michael Neuffer wrote: > > >This is not necessary. gcc 2.8 includes the pentium optimizations > >from pgcc. > > All of it? No not all, they took a stable subset. Mike -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Michael Neuffer wrote: >This is not necessary. gcc 2.8 includes the pentium optimizations >from pgcc. All of it? My impression from the pgcc FAQ at http://www.goof.com/ was that only some optimizations (mostly instruction scheduling) will be taken from pgcc. The rather active pgcc development at the moment (large patch files, lots of code reorganization) seems to indicate so. -- Thomas Koenig, [EMAIL PROTECTED], [EMAIL PROTECTED] The joy of engineering is to find a straight line on a double logarithmic diagram. -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On Sun, 1 Jun 1997, Thomas Koenig wrote: > >I don't think it does any optimization at all for pentium. > > Correct. Of course, there's the experimental pgcc (http://www.goof.com/, > if anybody wants to look). > > I'd like to pack this up and stuff it into experimental, if I had a > little more time *sigh*. This is not necessary. gcc 2.8 includes the pentium optimizations from pgcc. My guess is that it won't take very long anymore until 2.8 gets released. HJL found a few more bugs and his patches for libc6/glibc2 are not integrated yet, but otherwise it seems pretty stable now. Mike -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
[EMAIL PROTECTED] (Kai Henningsen) writes: > [EMAIL PROTECTED] (Christian Schwarz) wrote on 01.06.97 in <[EMAIL > PROTECTED]>: > > > Where is the arch specification string used, i.e. what will break if we > > change it to be "i386-linux" on intel systems? > > I'm not competent enough to answer this. Anything tightly integrated with > gcc, but is there anything that doesn't break already when the version > numbers don't match exactly? If i486-linux were changed to i386-linux then I would have to repackage Checker and reupload it. This would take a couple hours but wouldn't be too difficult. However, I don't see any point to the change. `A foolish consistency is the hobgoblin of little minds.' --Emerson -- Ben Pfaff <[EMAIL PROTECTED]> 12167 Airport Rd, DeWitt MI 48820, USA *Note*: New PGP key available at http://www.msu.edu/user/pfaffben/pgp.html -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
[EMAIL PROTECTED] (Christian Schwarz) wrote on 01.06.97 in <[EMAIL PROTECTED]>: > Where is the arch specification string used, i.e. what will break if we > change it to be "i386-linux" on intel systems? I'm not competent enough to answer this. Anything tightly integrated with gcc, but is there anything that doesn't break already when the version numbers don't match exactly? MfG Kai -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Rob Browning wrote: > > Galen Hazelwood <[EMAIL PROTECTED]> writes: > > > I think it also chooses some instructions differently for a 486, and > > these choices are also good on the pentium. That's why, when building > > binaries for my use, I use -m486 but add flags which turn off the > > alignment. > > Right, I had heard that these were reasonably good flags for the pentium: > > -m486 -malign-loops=2 -malign-jumps=2 -malign-functions=2 > Those are my personal favorites, at least. Best we can do until 2.8. --Galen -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Galen Hazelwood <[EMAIL PROTECTED]> writes: > I think it also chooses some instructions differently for a 486, and > these choices are also good on the pentium. That's why, when building > binaries for my use, I use -m486 but add flags which turn off the > alignment. Right, I had heard that these were reasonably good flags for the pentium: -m486 -malign-loops=2 -malign-jumps=2 -malign-functions=2 -- Rob -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On 1 Jun 1997, Guy Maor wrote: > Galen Hazelwood <[EMAIL PROTECTED]> writes: > > > Perhaps. Anybody have any serious arguments? I think the reason we > > configure gcc as i486 is so it automatically optimizes for the 486; it's > > a good middle ground. > > I think the only optimization gcc 2.7.* does for i486 is instruction > alignment. The Pentium has a better fetch unit so doesn't need any > alignment (it never incurs a misfetch penalty) so optimizing for i486 > will at least give some code bloat. > > I don't think it does any optimization at all for pentium. No not in 2.7.x, but there are noticable differences for P5 and P6 in gcc 2.8. Mike -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On Sun, 1 Jun 1997, Galen Hazelwood wrote: > Raul Miller wrote: > > > > On May 31, Galen Hazelwood wrote > > > Perhaps. Anybody have any serious arguments? I think the reason we > > > configure gcc as i486 is so it automatically optimizes for the 486; it's > > > a good middle ground. > > > > If I remember right, configuring for pentium leaves an executable > > that might not run on 386 or 486. > > > > If gcc actually _did_ any pentium optimizations, that might be true. > Unfortunately, we don't get that until 2.8. (still waiting...) On every compiler I have used they don't actually use pentium instructions. The optimization is soley from choosing order and instructions based on clock cycle considerations and with pentiums, pipelining too. So code optimized for the pentium will run on a 386, but the timings will not be as good for code optimized for the 386. Now, if GCC is weird and decided to use 486, P5 or P6 specific instructions (are there any new general use ones in the P5?) then this might be a problem. BTW, anyone know when 2.8 is due? Jason -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On 1 Jun 1997, Kai Henningsen wrote: > [EMAIL PROTECTED] (Galen Hazelwood) wrote on 31.05.97 in <[EMAIL PROTECTED]>: > > > Christian Schwarz wrote: > > > > > > On Thu, 29 May 1997, Galen Hazelwood wrote: > > > > (Don't ask me what the historical reasons are, though. I might start to > > > > whimper...) > > > > > > Sorry, but I couldn't resist :-) What are the reasons? > > > > I don't know. That's why I whimper... > > > > > If we make this policy, we should have some real arguments for it! Since > > > we use "i386" in all our file names and since Debian actually runs on > > > 386SX and higher I don't see why we should label this "i486". > > > > Perhaps. Anybody have any serious arguments? I think the reason we > > configure gcc as i486 is so it automatically optimizes for the 486; it's > > a good middle ground. > > That may be the reason that Linux gcc is usually configured that way. > > The reason Debian does it, AFAIR, is simply to be compatible with most of > the Linux world. Where is the arch specification string used, i.e. what will break if we change it to be "i386-linux" on intel systems? Thanks, Chris -- _,, Christian Schwarz / o \__ [EMAIL PROTECTED], [EMAIL PROTECTED], ! ___; [EMAIL PROTECTED], [EMAIL PROTECTED] \ / \\\__/ !PGP-fp: 8F 61 EB 6D CF 23 CA D7 34 05 14 5C C8 DC 22 BA \ / http://fatman.mathematik.tu-muenchen.de/~schwarz/ -.-.,---,-,-..---,-,-.,.-.- "DIE ENTE BLEIBT DRAUSSEN!" -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Guy Maor wrote: > I think the only optimization gcc 2.7.* does for i486 is instruction > alignment. The Pentium has a better fetch unit so doesn't need any > alignment (it never incurs a misfetch penalty) so optimizing for i486 > will at least give some code bloat. I think it also chooses some instructions differently for a 486, and these choices are also good on the pentium. That's why, when building binaries for my use, I use -m486 but add flags which turn off the alignment. > I don't think it does any optimization at all for pentium. True. 2.8 will have some stuff from pgcc, but not everything. --Galen -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Guy Maor wrote: [gcc 2.7.2] >I don't think it does any optimization at all for pentium. Correct. Of course, there's the experimental pgcc (http://www.goof.com/, if anybody wants to look). I'd like to pack this up and stuff it into experimental, if I had a little more time *sigh*. -- Thomas Koenig, [EMAIL PROTECTED], [EMAIL PROTECTED] The joy of engineering is to find a straight line on a double logarithmic diagram. -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Galen Hazelwood <[EMAIL PROTECTED]> writes: > Perhaps. Anybody have any serious arguments? I think the reason we > configure gcc as i486 is so it automatically optimizes for the 486; it's > a good middle ground. I think the only optimization gcc 2.7.* does for i486 is instruction alignment. The Pentium has a better fetch unit so doesn't need any alignment (it never incurs a misfetch penalty) so optimizing for i486 will at least give some code bloat. I don't think it does any optimization at all for pentium. Guy -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
[EMAIL PROTECTED] (Galen Hazelwood) wrote on 31.05.97 in <[EMAIL PROTECTED]>: > Christian Schwarz wrote: > > > > On Thu, 29 May 1997, Galen Hazelwood wrote: > > > (Don't ask me what the historical reasons are, though. I might start to > > > whimper...) > > > > Sorry, but I couldn't resist :-) What are the reasons? > > I don't know. That's why I whimper... > > > If we make this policy, we should have some real arguments for it! Since > > we use "i386" in all our file names and since Debian actually runs on > > 386SX and higher I don't see why we should label this "i486". > > Perhaps. Anybody have any serious arguments? I think the reason we > configure gcc as i486 is so it automatically optimizes for the 486; it's > a good middle ground. That may be the reason that Linux gcc is usually configured that way. The reason Debian does it, AFAIR, is simply to be compatible with most of the Linux world. MfG Kai -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Raul Miller wrote: > > On May 31, Galen Hazelwood wrote > > Perhaps. Anybody have any serious arguments? I think the reason we > > configure gcc as i486 is so it automatically optimizes for the 486; it's > > a good middle ground. > > If I remember right, configuring for pentium leaves an executable > that might not run on 386 or 486. > If gcc actually _did_ any pentium optimizations, that might be true. Unfortunately, we don't get that until 2.8. (still waiting...) --Galen -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On May 31, Galen Hazelwood wrote > Perhaps. Anybody have any serious arguments? I think the reason we > configure gcc as i486 is so it automatically optimizes for the 486; it's > a good middle ground. If I remember right, configuring for pentium leaves an executable that might not run on 386 or 486. -- Raul -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Christian Schwarz wrote: > > On Thu, 29 May 1997, Galen Hazelwood wrote: > > (Don't ask me what the historical reasons are, though. I might start to > > whimper...) > > Sorry, but I couldn't resist :-) What are the reasons? I don't know. That's why I whimper... > If we make this policy, we should have some real arguments for it! Since > we use "i386" in all our file names and since Debian actually runs on > 386SX and higher I don't see why we should label this "i486". Perhaps. Anybody have any serious arguments? I think the reason we configure gcc as i486 is so it automatically optimizes for the 486; it's a good middle ground. > Does someone know if it makes any troubles if we ommit the (ugly) > "unknown" in "i386-linux"? Not as far as I know. --Galen -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On Thu, 29 May 1997, Galen Hazelwood wrote: > Christian Schwarz wrote: > > > > Next step: GNU's "configure" utility. I thought that we had agreed on > > using > > i386-unknown-linux > > (and similar for the other architectures), but then I had just discovered > > that GCC uses > > /usr/lib/gcc-lib/i486-linux/2.7.2.1/ > > ^^ > > > > (first it says "i486" instead of "i386", second it ommits the "unknown"). > > > > Yes, for hysterical raisins, we use i486 instead of i386. > systems, these are identical. On Intel, they happen to be different. > We leave out the "unknown" becuase A) it's ugly, and B) nobody cares who > built your silly pc-clone box anyway. They're all the same, except for > the ways in which they are different. :) > > (Don't ask me what the historical reasons are, though. I might start to > whimper...) Sorry, but I couldn't resist :-) What are the reasons? If we make this policy, we should have some real arguments for it! Since we use "i386" in all our file names and since Debian actually runs on 386SX and higher I don't see why we should label this "i486". Does someone know if it makes any troubles if we ommit the (ugly) "unknown" in "i386-linux"? Thanks, Chris -- _,, Christian Schwarz / o \__ [EMAIL PROTECTED], [EMAIL PROTECTED], ! ___; [EMAIL PROTECTED], [EMAIL PROTECTED] \ / \\\__/ !PGP-fp: 8F 61 EB 6D CF 23 CA D7 34 05 14 5C C8 DC 22 BA \ / http://fatman.mathematik.tu-muenchen.de/~schwarz/ -.-.,---,-,-..---,-,-.,.-.- "DIE ENTE BLEIBT DRAUSSEN!" -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On Thu, 29 May 1997, Christian Schwarz wrote: > Hi folks! > > As I'm working on a new Policy I'm handling the request to include a > policy for "correct architecture spec strings". However, I've discovered > _several_ threads here on debian-devel without any (obvious) results. > > Is it correct, that we are currently working on ports to the following > platforms (the abbrevs should be the ones that dpkg is using in the file > names): > i386 > alpha > arm > m68k > powerpc > sparc > Are these correct (i.e. not ppc) and is this list complete? > > Next step: GNU's "configure" utility. I thought that we had agreed on > using > i386-unknown-linux > (and similar for the other architectures), but then I had just discovered > that GCC uses > /usr/lib/gcc-lib/i486-linux/2.7.2.1/ > ^^ > > (first it says "i486" instead of "i386", second it ommits the "unknown"). No, i486-linux is the alias we should use. Latest configure tools will convert this to i486-pc-linux-gnu! But, all programs should install using the alias, not the full name. Dpkg calls it "i386", but the autoconf string should be i486. In general, the alias given to configure should be xxx-linux. -- Tom Lees <[EMAIL PROTECTED]>http://www.lpsg.demon.co.uk/ PGP ID 87D4D065, fingerprint 2A 66 86 9D 02 4D A6 1E B8 A2 17 9D 4F 9B 89 D6 finger [EMAIL PROTECTED] for full public key (also available on keyservers) -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On Thu, 29 May 1997, Vincent Renardias wrote: > > On Thu, 29 May 1997, Christian Schwarz wrote: > > > Is it correct, that we are currently working on ports to the following > > platforms (the abbrevs should be the ones that dpkg is using in the file > > names): > > i386 > > alpha > > arm > > m68k > > powerpc > > sparc > > Are these correct (i.e. not ppc) and is this list complete? > > Correct for powerpc. > Didn't know someone was working on ARM(who?) We are going to have debian/ARM? :> I know there is a Linux port of some form for some of the ARM chips, Esp the StrongARM... Anyone know how well they run linux? Jason Mmm, 50$ 200Mips RISC processor.. I want 6! -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
Christian Schwarz wrote: > > Next step: GNU's "configure" utility. I thought that we had agreed on > using > i386-unknown-linux > (and similar for the other architectures), but then I had just discovered > that GCC uses > /usr/lib/gcc-lib/i486-linux/2.7.2.1/ > ^^ > > (first it says "i486" instead of "i386", second it ommits the "unknown"). > Yes, for hysterical raisins, we use i486 instead of i386. The difference is reflected in the results you get form dpkg --print-architecture and dpkg --print-gnu-build-architecture. On most systems, these are identical. On Intel, they happen to be different. We leave out the "unknown" becuase A) it's ugly, and B) nobody cares who built your silly pc-clone box anyway. They're all the same, except for the ways in which they are different. :) (Don't ask me what the historical reasons are, though. I might start to whimper...) --Galen -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
Re: RFC: Policy for arch specs
On Thu, 29 May 1997, Christian Schwarz wrote: > Is it correct, that we are currently working on ports to the following > platforms (the abbrevs should be the ones that dpkg is using in the file > names): > i386 > alpha > arm > m68k > powerpc > sparc > Are these correct (i.e. not ppc) and is this list complete? Correct for powerpc. Didn't know someone was working on ARM(who?) Cordialement, -- - ** Linux ** +---+ ** WAW ** - - [EMAIL PROTECTED] | RENARDIAS Vincent | [EMAIL PROTECTED] - - Debian/GNU Linux +---+ http://www.waw.com/ - - http://www.debian.org/ |WAW (33) 4 91 81 21 45 - --- -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
RFC: Policy for arch specs
Hi folks! As I'm working on a new Policy I'm handling the request to include a policy for "correct architecture spec strings". However, I've discovered _several_ threads here on debian-devel without any (obvious) results. Is it correct, that we are currently working on ports to the following platforms (the abbrevs should be the ones that dpkg is using in the file names): i386 alpha arm m68k powerpc sparc Are these correct (i.e. not ppc) and is this list complete? Next step: GNU's "configure" utility. I thought that we had agreed on using i386-unknown-linux (and similar for the other architectures), but then I had just discovered that GCC uses /usr/lib/gcc-lib/i486-linux/2.7.2.1/ ^^ (first it says "i486" instead of "i386", second it ommits the "unknown"). (The reason we didn't choose i386-debian-linux is that we wanted our utilities to be compatible with other distributions.) Any comments, please. Thanks, Chris -- Christian Schwarz [EMAIL PROTECTED], [EMAIL PROTECTED], Don't know Perl? [EMAIL PROTECTED], [EMAIL PROTECTED] Visit PGP-fp: 8F 61 EB 6D CF 23 CA D7 34 05 14 5C C8 DC 22 BA http://www.perl.com http://fatman.mathematik.tu-muenchen.de/~schwarz/ -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
