Re: gcc -Os optimisation broken (RELENG_4)
David O'Brien wrote/schrieb (Saturday, March 18, 2000): | On Sat, Mar 18, 2000 at 03:18:45AM +0100, Thomas Köllmann wrote: | | Perhaps this is a bit off topic, but can the pentium optimisations be | | used for AMD K6 processors? | | I did a `make world' yesterday with | CFLAGS= -O2 -pipe -march=pentium | COPTFLAGS= -O2 -pipe -march=pentium | ..snip.. | If it doesn't I'll probably try `-03 -pipe -march=pentium' come next | | What are people hoping to get by doing this? Are you actually doing a | scientific performance evaluation between the various optimization | options??? This is just playing, the machine I was talking about has it's backups in order and can afford downtime; I mentioned that already, and I was only answering somebody's question. | Are are people just being macho, and thinking they are | getting all this non-existent performance increase? You _are_ feeling strong about this, aren't you? :-) | "-O" is the only globally safe optimization on FreeBSD. -O2, etc.. | causes various problems for various people in various ports, and parts of | /usr/src/. If people are using these options just for fun, that is fine, Yes, just for fun, David, just for fun. | BUT if you experience *any* problems with compiling using -O2, etc.. | don't bug this list -- go bug the GCC people. Are they a bunch of machos themselves? :-) Thanks for your point of view. Gruß - Thomas -- Walking to the car, she takes his hand and puts it, for a moment, lightly between her moving legs. Roger's heart grows erect, and comes. That's really how it feels. -- Thomas Pynchon, Gravity's Rainbow # PGP key sent on request / PGP key auf Wunsch per e-mail To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
I wrote/schrieb (Saturday, March 18, 2000): | R Joseph Wright wrote/schrieb (Friday, March 17, 2000): | | | In contrast, I've been using -Os -march=pentium during the last three | | months for buildworld and the kernel. Never had problems whatsoever. | | | | Perhaps this is a bit off topic, but can the pentium optimisations be used | | for AMD K6 processors? | | I did a `make world' yesterday with | CFLAGS= -O2 -pipe -march=pentium | COPTFLAGS= -O2 -pipe -march=pentium | (inspired by recent mention of optimizing on this list) on my | AMD-K6(tm) 3D processor (350.80-MHz 586-class CPU) | desktop machine. It still runs, but I'll have yet to see whether | this will cause any suspicious behaviour. [...] I've just learned that there is a `-march=k6' even though the man page for gcc-2.95.1 does not mention it... Gruß - Thomas -- Den Wein, den listigen Tröster bat ich oft, | - Charles Einmal das Schrecknis, das mich quält, zu stillen, | Baudelaire, Jedoch er schärft den Sinn, statt zu verhüllen | Die Blutquelle # PGP key sent on request / PGP key auf Wunsch per e-mail To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Thomas Köllmann wrote: I wrote/schrieb (Saturday, March 18, 2000): | R Joseph Wright wrote/schrieb (Friday, March 17, 2000): | | | In contrast, I've been using -Os -march=pentium during the last three | | months for buildworld and the kernel. Never had problems whatsoever. | | | | Perhaps this is a bit off topic, but can the pentium optimisations be used | | for AMD K6 processors? | | I did a `make world' yesterday with | CFLAGS= -O2 -pipe -march=pentium | COPTFLAGS= -O2 -pipe -march=pentium | (inspired by recent mention of optimizing on this list) on my | AMD-K6(tm) 3D processor (350.80-MHz 586-class CPU) | desktop machine. It still runs, but I'll have yet to see whether | this will cause any suspicious behaviour. [...] I've just learned that there is a `-march=k6' even though the man page for gcc-2.95.1 does not mention it... GNU folks are very reluctant in respect of updating man pages (they think that man pages is obsolete feature). If you would look into gcc man page thoroughly you could find following disclaimer in its very beginning: This man page is not kept up to date except when volun- teers want to maintain it. If you find a discrepancy between the man page and the software, please check the Info file, which is the authoritative documentation. Therefore if you want to learn about new options and optimisations than it would be better to read info pages instead (info gcc). -Maxim To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
On Sat, Mar 18, 2000 at 03:18:45AM +0100, Thomas Köllmann wrote: | Perhaps this is a bit off topic, but can the pentium optimisations be used | for AMD K6 processors? I did a `make world' yesterday with CFLAGS= -O2 -pipe -march=pentium COPTFLAGS= -O2 -pipe -march=pentium ..snip.. If it doesn't I'll probably try `-03 -pipe -march=pentium' come next What are people hoping to get by doing this? Are you actually doing a scientific performance evaluation between the various optimization options??? Are are people just being macho, and thinking they are getting all this non-existent performance increase? "-O" is the only globally safe optimization on FreeBSD. -O2, etc.. causes various problems for various people in various ports, and parts of /usr/src/. If people are using these options just for fun, that is fine, BUT if you experience *any* problems with compiling using -O2, etc.. don't bug this list -- go bug the GCC people. -- -- David([EMAIL PROTECTED]) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
David O'Brien wrote: On Sat, Mar 18, 2000 at 03:18:45AM +0100, Thomas Köllmann wrote: If it doesn't I'll probably try `-03 -pipe -march=pentium' come next What are people hoping to get by doing this? Are you actually doing a scientific performance evaluation between the various optimization options??? Are are people just being macho, and thinking they are getting all this non-existent performance increase? It's probably more of a "placebo effect", which makes you think your are getting a big boost in performance. I'll admit that I've never seen a whole order or magnitude increase in performance between -O and -mpentium-O3, or whatever - it probably gives you boosts here and there. Optimization is pretty good, as I've found out, with plain-jane -O. Beyond that, I think your performance gains are minimal. And yes, I think it's really macho to be usin' hopped-up CFLAGS, like -march=pentium -Os -pipe. I feel really studly doing this. :-) We should do a survey, and find out what the guys use for CFLAGS and COPTFLAGS and compare them to what the female users are using. That would be interesting. I don't know, -O seems to be doing some pretty decent optimization, from what I've seen of the assembler output. I guess you're right in that all higher optimization are prividing marginal performance gains. From what I gather, the i386 arch. is not very sensitive to optimization from gcc. You'll see minor boosts here and there, but that's about it. Typically, guys that are diehard football fans like to use -mpentium -O3 -pipe for building world and kernel. - Donn To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Donn Miller wrote: It's probably more of a "placebo effect", which makes you think your are getting a big boost in performance. I'll admit that I've never seen a whole order or magnitude increase in performance between -O and -mpentium-O3, or whatever - it probably gives you boosts here and there. Optimization is pretty good, as I've found out, with plain-jane -O. Beyond that, I think your performance gains are minimal. And yes, I think it's really macho to be usin' hopped-up CFLAGS, like -march=pentium -Os -pipe. I feel really studly doing this. :-) Damn straight! But you ain't seen nothing yet... We should do a survey, and find out what the guys use for CFLAGS and COPTFLAGS and compare them to what the female users are using. That would be interesting. Let's get the ball rolling then, shall we? CFLAGS/COPTFLAGS for day-to-day compiles (kernel, X, world, etc.): -O3 -march=pentiumpro -malign-double -fomit-frame-pointer -funroll-loops -fstrict-aliasing -pipe CFLAGS when I *REALLY* feel the need for speed (lame, mpg123, xmame, etc.): -O3 -march=pentiumpro -malign-double -malign-loops=4 -malign-jumps=4 -malign-functions=4 -fomit-frame-pointer -ffast-math -funroll-loops -fstrict-aliasing -pipe -O2 used to be my switch of choice so the -funroll-loops flag in the above strings may be redundant (can't remember). Considering, though, that the length of one's CFLAGS variable is directly proportional to studliness, I have little choice but to leave it there ;-) Can't say that I care for football, though. - Donn -jake [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
On Thu, Mar 16, 2000 at 10:09:37PM -0800, Doug Barton wrote: Donn Miller wrote: Doug Barton wrote: Hmm... If I have a PII (Actually celeron 300A) or a PIII, which is better, 'pentium' or 'pentiumpro'? I would think the latter, but I've learned not to assume where gcc is concerned. I think that 'pentium' would result in code that isn't as optimized as 'pentiumpro', but I've heard that 'pentium' has a lot less problems. Also, I have heard conflicting reports as to whether compiling the kernel/world with optimisations is a good thing. Anyone care to (re)open that can of worms? I compile my kernel/world with -mpentium -O3 -pipe. The only problem I've seen so far were spurious random reboots that would occur about 2-3 times a month. But, that was last summer, and hasn't happened since. Something else must have been the culprit. (Maybe -current wasn't as stable last summer.) With the aforementioned CFLAGS, I have a pretty reliable and stable system. I've heard that -mpentiumpro can be pretty buggy, and it can actually result in slower code than -mpentium for certain pentium types. I trust plain -mpentium, as it has been very reliable for me, except for some compile-time errors caused by the optimization (Qt). In the interests of providing another datapoint, I tried my old, boring P5 machine, and with -Os -march=pentium buildworld bombed trying to compile cc1plus in the build tools phase. Backing off to -O worked. The kernel was ok with -Os -march=pentium. In contrast, I've been using -Os -march=pentium during the last three months for buildworld and the kernel. Never had problems whatsoever. - Sascha To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
On 16 Mar, Doug Barton wrote:
In the interests of providing another datapoint, I tried my old, boring
P5 machine, and with -Os -march=pentium buildworld bombed trying to
compile cc1plus in the build tools phase. Backing off to -O worked. The
kernel was ok with -Os -march=pentium.
As it seems everyone is posting his/her C{,OPT}FLAGS:
I'm using
-Os -march=pentiumpro -pipe -Wall -funroll-loops -fschedule-insns2
since months (~ a half year ore more) without a problem (at least I
didn't notice one) on a Celeron.
Bye,
Alexander.
--
The dark ages were caused by the Y1K problem.
http://www.Leidinger.net Alexander+Home @ Leidinger.net
Key fingerprint = 7423 F3E6 3A7E B334 A9CC B10A 1F5F 130A A638 6E7E
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Maxim Sobolev wrote: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. After some digging into the sources with gdb I found that fault came from dereferencing NULL pointer somewhere in the DNS query procedures. I've tracked the source of this pointer and found that the function rfc1035QuestionPack (rfc1035.c) called from rfc1035BuildAQuery receives NULL pointer instead of the supplied hostname as a 3rd argument. Following is two debugging sessions with squid compiled with -Os and -O (faulty call is in the end of the output): Well... where is "name" being set? That would help. -- Daniel C. Sobral(8-DCS) [EMAIL PROTECTED] [EMAIL PROTECTED] One Unix to rule them all, One Resolver to find them, One IP to bring them all and in the zone bind them. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Maxim Sobolev wrote: Well... where is "name" being set? That would help. It is not clear what do you mean, please explain. "name" is the name of the variable that is passed as NULL when compiled with -Os. In the code trace you posted, we do not see any reference to this variable up to the point where the function is called. -- Daniel C. Sobral(8-DCS) [EMAIL PROTECTED] [EMAIL PROTECTED] One Unix to rule them all, One Resolver to find them, One IP to bring them all and in the zone bind them. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
In contrast, I've been using -Os -march=pentium during the last three months for buildworld and the kernel. Never had problems whatsoever. Perhaps this is a bit off topic, but can the pentium optimisations be used for AMD K6 processors? To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
"Daniel C. Sobral" wrote: Maxim Sobolev wrote: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. After some digging into the sources with gdb I found that fault came from dereferencing NULL pointer somewhere in the DNS query procedures. I've tracked the source of this pointer and found that the function rfc1035QuestionPack (rfc1035.c) called from rfc1035BuildAQuery receives NULL pointer instead of the supplied hostname as a 3rd argument. Following is two debugging sessions with squid compiled with -Os and -O (faulty call is in the end of the output): Well... where is "name" being set? That would help. It is not clear what do you mean, please explain. -Maxim To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
R Joseph Wright wrote/schrieb (Friday, March 17, 2000): | In contrast, I've been using -Os -march=pentium during the last three | months for buildworld and the kernel. Never had problems whatsoever. | | Perhaps this is a bit off topic, but can the pentium optimisations be used | for AMD K6 processors? I did a `make world' yesterday with CFLAGS= -O2 -pipe -march=pentium COPTFLAGS= -O2 -pipe -march=pentium (inspired by recent mention of optimizing on this list) on my AMD-K6(tm) 3D processor (350.80-MHz 586-class CPU) desktop machine. It still runs, but I'll have yet to see whether this will cause any suspicious behaviour. If it doesn't I'll probably try `-03 -pipe -march=pentium' come next `make world' time, but this all is about a machine that has it's backup mechanisms (hopefully) and can afford a little downtime. Gruß - Thomas -- On the way downtown I stopped at a bar and had a couple of double Scotches. They didn't do me any good. All they did was make me think of Silver-Wig, and I never saw her again. -- Raymond Chandler, The Big Sleep # PGP key sent on request / PGP key auf Wunsch per e-mail To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
I think that 'pentium' would result in code that isn't as optimized as 'pentiumpro', but I've heard that 'pentium' has a lot less problems. What??? 'pentiumpro' code isn't going to be very optimized for a Pentium (if it even runs at all). I've heard that -mpentiumpro can be pretty buggy, and it can actually result in slower code than -mpentium for certain pentium types. Yea like the original P5 Pentiums. You should match the command line with your actual machine if you are going to use these options. -- -- David([EMAIL PROTECTED]) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Doug Barton [EMAIL PROTECTED] wrote: Hmm... If I have a PII (Actually celeron 300A) or a PIII, which is better, 'pentium' or 'pentiumpro'? I would think the latter, but I've I have to admit that I kind of lost track of Intel's Pentium du jour offerings after the PPro, but I think PII and PIII use i686 cores or at least something closest to i686, so I'd use -mpentiumpro there. I don't pretend to have any idea what's appropriate for the various AMD/Cyrix/IDT/etc processors. The machines where I use -mpentium and -mpentiumpro, respectively, are an actual Intel Pentium and a (dual) Intel Pentium Pro. Yes, there are people out there who don't buy a new machine each quarter. Also, I have heard conflicting reports as to whether compiling the kernel/world with optimisations is a good thing. Anyone care to (re)open that can of worms? No new worms in there, I suspect. -- Christian "naddy" Weisgerber [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
David O'Brien [EMAIL PROTECTED] wrote: What??? 'pentiumpro' code isn't going to be very optimized for a Pentium (if it even runs at all). According to the gcc(1) man page, -mpentiumpro is synonymous to -mcpu=pentiumpro, which only affects instruction scheduling but not the actual instruction set used (for that, use -march=...). So it certainly should run. If you are aware that the man page is wrong in this respect, please tell us! -- Christian "naddy" Weisgerber [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
At 01:42 PM 3/16/00 +0100, Christian Weisgerber wrote: David O'Brien [EMAIL PROTECTED] wrote: What??? 'pentiumpro' code isn't going to be very optimized for a Pentium (if it even runs at all). According to the gcc(1) man page, -mpentiumpro is synonymous to -mcpu=pentiumpro, which only affects instruction scheduling but not the actual instruction set used (for that, use -march=...). So it certainly should run. If you are aware that the man page is wrong in this respect, please tell us! Wondering why one would use -mcpu and not -march. If the code runs only on Celerons, PII's, and PIII's why would one *not* use -march. I'm curious about (possible) breakages with -mcpu or -march compared to -Ox settings which seem to break things more often than -O. Only ask, since -Ox and individual flags (rather than the mulititude added going from -O to -O2) are used far more often. Jeff Mountin - [EMAIL PROTECTED] Systems/Network Administrator FreeBSD - the power to serve To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Donn Miller wrote: Doug Barton wrote: Hmm... If I have a PII (Actually celeron 300A) or a PIII, which is better, 'pentium' or 'pentiumpro'? I would think the latter, but I've learned not to assume where gcc is concerned. I think that 'pentium' would result in code that isn't as optimized as 'pentiumpro', but I've heard that 'pentium' has a lot less problems. Also, I have heard conflicting reports as to whether compiling the kernel/world with optimisations is a good thing. Anyone care to (re)open that can of worms? I compile my kernel/world with -mpentium -O3 -pipe. The only problem I've seen so far were spurious random reboots that would occur about 2-3 times a month. But, that was last summer, and hasn't happened since. Something else must have been the culprit. (Maybe -current wasn't as stable last summer.) With the aforementioned CFLAGS, I have a pretty reliable and stable system. I've heard that -mpentiumpro can be pretty buggy, and it can actually result in slower code than -mpentium for certain pentium types. I trust plain -mpentium, as it has been very reliable for me, except for some compile-time errors caused by the optimization (Qt). In the interests of providing another datapoint, I tried my old, boring P5 machine, and with -Os -march=pentium buildworld bombed trying to compile cc1plus in the build tools phase. Backing off to -O worked. The kernel was ok with -Os -march=pentium. Hope someone is finding this useful, Doug -- "While the future's there for anyone to change, still you know it seems, it would be easier sometimes to change the past" - Jackson Browne, "Fountain of Sorrow" To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
On Thu, 16 Mar 2000, Jeffrey J. Mountin wrote: Wondering why one would use -mcpu and not -march. If the code runs only on Celerons, PII's, and PIII's why would one *not* use -march. I'm curious about (possible) breakages with -mcpu or -march compared to -Ox settings which seem to break things more often than -O. Only ask, since -Ox and individual flags (rather than the mulititude added going from -O to -O2) are used far more often. Eager to fire my new-found gun in the direction of my feet I built world and kernel last night with -0s -march=pentium. So far so good (although I haven't given it a real workout yet). Now that I now 'pentiumpro' should be a better choice, I'll give that a whirl tonight. After reading the man page I had to agree with your point that -march seemed like a better option, and I don't have cross-platform issues to deal with here. Doug -- "While the future's there for anyone to change, still you know it seems, it would be easier sometimes to change the past" - Jackson Browne, "Fountain of Sorrow" To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
In the last episode (Mar 15), Maxim Sobolev said: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. After some digging into the sources with gdb I found This is just a 'me too'. I get it with -O2 (-Os implies -O2, so it's probably the same problem). -- Dan Nelson [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Dan Nelson wrote: In the last episode (Mar 15), Maxim Sobolev said: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. After some digging into the sources with gdb I found This is just a 'me too'. I get it with -O2 (-Os implies -O2, so it's probably the same problem). I've noticed various compile-time optimization bugs as well. For example, I tried building Qt with -mpentium -O3 -pipe, and somewhere during the build, I get "Internal compiler error." Falling back to the stock optimization levels of -O2 fixed this. - Donn To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Dan Nelson wrote: In the last episode (Mar 15), Maxim Sobolev said: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. After some digging into the sources with gdb I found This is just a 'me too'. I get it with -O2 (-Os implies -O2, so it's probably the same problem). Any comments from the "compiler gurus"? -Maxim To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
On Wed, Mar 15, 2000 at 10:51:55AM -0600, Dan Nelson wrote: I get it with -O2 (-Os implies -O2, so it's probably the same problem). Not quite. -0s == all the -O2 optimizations that do not increase code size. -Os can also perform other optimizations not part of -O2 that decrease code size. The -Os == -O2 only tells you how "risky" in optimizing -Os is willing to be. -- -- David([EMAIL PROTECTED]) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
In the last episode (Mar 15), David O'Brien said: On Wed, Mar 15, 2000 at 10:51:55AM -0600, Dan Nelson wrote: I get it with -O2 (-Os implies -O2, so it's probably the same problem). Not quite. -0s == all the -O2 optimizations that do not increase code size. -Os can also perform other optimizations not part of -O2 that decrease code size. The -Os == -O2 only tells you how "risky" in optimizing -Os is willing to be. Too risky, apparently :) Maxim: It looks like you've done quite a big of debugging already; can you get this bug to appear in a small piece of code? I'm sure the gcc developers would be able to fix the problem pretty quickly if it's easily reproducable. -- Dan Nelson [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Maxim Sobolev [EMAIL PROTECTED] wrote: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. Which brings us back to the popular topic of which optimization levels are reliable. Building 5.0-CURRENT with "-Os -mpentium" failed for me with gas complaining (unknown opcode I think, but don't hold me to it). I went back to "-O -mpentium" and that worked fine. On another machine, "-O2 -mpentiumpro" worked, too. -- Christian "naddy" Weisgerber [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: gcc -Os optimisation broken (RELENG_4)
Christian Weisgerber wrote: Maxim Sobolev [EMAIL PROTECTED] wrote: I've just upgraded my production server to the 4.0-RELEASE and found that squid23 when compiled with -Os option dying with signal 11 on each attempt to load page. When I recompiled it with -O fault disappeared. Which brings us back to the popular topic of which optimization levels are reliable. Building 5.0-CURRENT with "-Os -mpentium" failed for me with gas complaining (unknown opcode I think, but don't hold me to it). I went back to "-O -mpentium" and that worked fine. On another machine, "-O2 -mpentiumpro" worked, too. Hmm... If I have a PII (Actually celeron 300A) or a PIII, which is better, 'pentium' or 'pentiumpro'? I would think the latter, but I've learned not to assume where gcc is concerned. Also, I have heard conflicting reports as to whether compiling the kernel/world with optimisations is a good thing. Anyone care to (re)open that can of worms? Thanks, Doug -- "While the future's there for anyone to change, still you know it seems, it would be easier sometimes to change the past" - Jackson Browne, "Fountain of Sorrow" To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
