gcc-3.1

bridged with qdn.public.qnxrtp.porting
Post Reply
Jörn Ihlenburg

gcc-3.1

Post by Jörn Ihlenburg » Sat May 25, 2002 12:02 am

Hi everybody,

I tried to compile gcc-3.1 for QNX/x86 but i got stuck with 'configure'.
Has somebody sucessfully compiled gcc-3.x.x for QNX?
If not, what steps have to be done to do this?
Many thanks in advance...

Joern

Kris Warkentin

Re: gcc-3.1

Post by Kris Warkentin » Mon May 27, 2002 5:38 pm

We haven't ported our stuff to gcc-3.1 yet. What you would have to do is
grab our 2.95.3 sources from qdn.qnx.com and apply those changes to the 3.1
branch. Internally, Graeme was taking a look at it and got it partially
working for x86 only but we haven't had time to finish it up yet. This
isn't on any schedule so don't hold your breath. As far as I know, we'll be
sticking with the 2.95 series for quite a while.

cheers,

Kris

"Jörn Ihlenburg" <ihlenburg@web.de> wrote in message
news:acmjk1$ipu$1@inn.qnx.com...
Hi everybody,

I tried to compile gcc-3.1 for QNX/x86 but i got stuck with 'configure'.
Has somebody sucessfully compiled gcc-3.x.x for QNX?
If not, what steps have to be done to do this?
Many thanks in advance...

Joern

Jörn Ihlenburg

Re: gcc-3.1

Post by Jörn Ihlenburg » Mon May 27, 2002 8:28 pm

Hi Kris,

sorry to hear that... I was trying to build a gcj-3.1, so I needed to
compile gcc-3.1. But as I see, this will become quite problematic. Since
porting gcj isn't on my shedule either (I was doing it just for fun :) ),
I'll leave it open until somone else (maybe you?) will port it in the
future. Just because I'm curious, why do you stick with 2.95? I know, nwer
isn't always better, but gcc-3.1 does have some major advantages...
Thank you very much for your reply anyway...

cheers

Joern



"Kris Warkentin" <kewarken@qnx.com> schrieb im Newsbeitrag
news:actqgm$eea$1@nntp.qnx.com...
We haven't ported our stuff to gcc-3.1 yet. What you would have to do is
grab our 2.95.3 sources from qdn.qnx.com and apply those changes to the
3.1
branch. Internally, Graeme was taking a look at it and got it partially
working for x86 only but we haven't had time to finish it up yet. This
isn't on any schedule so don't hold your breath. As far as I know, we'll
be
sticking with the 2.95 series for quite a while.

cheers,

Kris

"Jörn Ihlenburg" <ihlenburg@web.de> wrote in message
news:acmjk1$ipu$1@inn.qnx.com...
Hi everybody,

I tried to compile gcc-3.1 for QNX/x86 but i got stuck with 'configure'.
Has somebody sucessfully compiled gcc-3.x.x for QNX?
If not, what steps have to be done to do this?
Many thanks in advance...

Joern



Kris Warkentin

Re: gcc-3.1

Post by Kris Warkentin » Mon May 27, 2002 9:00 pm

"Jörn Ihlenburg" <ihlenburg@web.de> wrote in message
news:acu45s$ln$1@inn.qnx.com...
Hi Kris,

sorry to hear that... I was trying to build a gcj-3.1, so I needed to
compile gcc-3.1. But as I see, this will become quite problematic. Since
porting gcj isn't on my shedule either (I was doing it just for fun :) ),
I'll leave it open until somone else (maybe you?) will port it in the
future. Just because I'm curious, why do you stick with 2.95? I know, nwer
Stability. We've got customers who rely on our toolchain so once things are
running fairly smoothly, we're reluctant to jump about too much. What we
will probably do is a test port to release for people (enthusiasts?) to play
with and see what everyone thinks. Our ultimate goal is to have our
toolchain synced with the gnu head branch so it may be that you will someday
be able to grab a gcc/gdb/binutils snapshot right off of ftp.gnu.org and
build our toolchain.
isn't always better, but gcc-3.1 does have some major advantages...
Some. And some disadvantages too. It's a heck of a lot slower on the
compile. ;-)
Thank you very much for your reply anyway...

cheers

Joern

Mario Charest

Re: gcc-3.1

Post by Mario Charest » Mon May 27, 2002 11:48 pm

Some. And some disadvantages too. It's a heck of a lot slower on the
compile. ;-)
I though 3.x was faster then 2.9.x. Oh well, lets just hope computer
processing
power will increase by the same factor that GCC speed decreases

Funny cause I just had a customer where complete build of their system
takes 4 hours on QNX4. I estimated it was going to take 12 hours to compile
on QNX6. If a "heck of a lot slower" means 2x slower that would mean
24 hours, ouch!
Thank you very much for your reply anyway...

cheers

Joern


Yuri D'Elia

Re: gcc-3.1

Post by Yuri D'Elia » Tue May 28, 2002 8:32 am

In article <acu6co$mi4$1@nntp.qnx.com>, Kris Warkentin wrote:
Some. And some disadvantages too. It's a heck of a lot slower on the
compile. ;-)
Oh boy :). Let's immagine gcc3 speed ported to qnx as it was for 2.95 :)

--
Wave++ (Yuri D'Elia)
Software Developer @ ubiest.com

Rennie Allen

Re: gcc-3.1

Post by Rennie Allen » Tue May 28, 2002 9:51 am

Kris Warkentin wrote:
Recently some of our guys were asked to try and optimize the performance of
Eclipse on RTP and in very short order were able to make it comparable to
Windows and Solaris.
Wow ! This is good news. Both in that mgmt at QSSL still views
self-hosted as enough of a priority to issue such an edict, and that I
am currently in the process of downloading it. Now if only we could
optimize the internet such that downloading a 640MB image didn't take so
long :-)

Rennie

Joern Ihlenburg

Re: gcc-3.1

Post by Joern Ihlenburg » Tue May 28, 2002 1:22 pm

Hi Kris,

"Kris Warkentin" <kewarken@qnx.com> schrieb im Newsbeitrag
news:acu6co$mi4$1@nntp.qnx.com...
Our ultimate goal is to have our
toolchain synced with the gnu head branch so it may be that you will
someday
be able to grab a gcc/gdb/binutils snapshot right off of ftp.gnu.org and
build our toolchain.
This would be so cool... Until then, I'll keep on dreaming :)
Many thanks...

Joern

Kris Warkentin

Re: gcc-3.1

Post by Kris Warkentin » Tue May 28, 2002 1:39 pm

"Mario Charest" <goto@nothingness.com> wrote in message
news:acuft6$847$1@inn.qnx.com...

Some. And some disadvantages too. It's a heck of a lot slower on the
compile. ;-)

I though 3.x was faster then 2.9.x. Oh well, lets just hope computer
processing
power will increase by the same factor that GCC speed decreases
Actually, I've been following the gcc mailing list and there are some
serious concerns, especially with the 3.1 series. OpenBSD folk are
complaining that it basically makes many of the older, slower platforms
unusable (4 days to compile the kernel when it used to take 18 hours, etc.)
Funny cause I just had a customer where complete build of their system
takes 4 hours on QNX4. I estimated it was going to take 12 hours to
compile
on QNX6. If a "heck of a lot slower" means 2x slower that would mean
24 hours, ouch!
Yes. However, many of the performance hits are us, not gcc. If you compile
the same project on the same machine with the same gcc, using Linux and
QNX6, you'll see significant differences. So there are some bottlenecks in
things like IO than need to be improved on our end as well.
Thank you very much for your reply anyway...

cheers

Joern




Kris Warkentin

Re: gcc-3.1

Post by Kris Warkentin » Tue May 28, 2002 1:41 pm

"Joern Ihlenburg" <joern.ihlenburg@bizerba.com> wrote in message
news:acvvjr$bsc$1@inn.qnx.com...
Hi Kris,

"Kris Warkentin" <kewarken@qnx.com> schrieb im Newsbeitrag
news:acu6co$mi4$1@nntp.qnx.com...
Our ultimate goal is to have our
toolchain synced with the gnu head branch so it may be that you will
someday
be able to grab a gcc/gdb/binutils snapshot right off of ftp.gnu.org and
build our toolchain.


This would be so cool... Until then, I'll keep on dreaming :)
The dream may be closer than you think - our binutils patch is currently in
FSF hands, hopefully just needing some tweaking (fingers crossed ;-) before
they accept it into their main branch.
Many thanks...

Joern


Mario Charest

Re: gcc-3.1

Post by Mario Charest » Tue May 28, 2002 2:17 pm

"Kris Warkentin" <kewarken@qnx.com> wrote in message
news:ad00tp$3md$1@nntp.qnx.com...
"Mario Charest" <goto@nothingness.com> wrote in message
news:acuft6$847$1@inn.qnx.com...


Some. And some disadvantages too. It's a heck of a lot slower on the
compile. ;-)

I though 3.x was faster then 2.9.x. Oh well, lets just hope computer
processing
power will increase by the same factor that GCC speed decreases

Actually, I've been following the gcc mailing list and there are some
serious concerns, especially with the 3.1 series. OpenBSD folk are
complaining that it basically makes many of the older, slower platforms
unusable (4 days to compile the kernel when it used to take 18 hours,
etc.)
Ouch!
Funny cause I just had a customer where complete build of their system
takes 4 hours on QNX4. I estimated it was going to take 12 hours to
compile
on QNX6. If a "heck of a lot slower" means 2x slower that would mean
24 hours, ouch!

Yes. However, many of the performance hits are us, not gcc. If you
compile
the same project on the same machine with the same gcc, using Linux and
QNX6, you'll see significant differences. So there are some bottlenecks
in
things like IO than need to be improved on our end as well.
That's encouraging, as it means to me you can do something about it ;-)

Thank you very much for your reply anyway...

cheers

Joern






Kris Warkentin

Re: gcc-3.1

Post by Kris Warkentin » Tue May 28, 2002 2:32 pm

"Mario Charest" <goto@nothingness.com> wrote in message
news:ad02pq$ec5$1@inn.qnx.com...
"Kris Warkentin" <kewarken@qnx.com> wrote in message
news:ad00tp$3md$1@nntp.qnx.com...
"Mario Charest" <goto@nothingness.com> wrote in message
news:acuft6$847$1@inn.qnx.com...


Funny cause I just had a customer where complete build of their
system
takes 4 hours on QNX4. I estimated it was going to take 12 hours to
compile
on QNX6. If a "heck of a lot slower" means 2x slower that would mean
24 hours, ouch!

Yes. However, many of the performance hits are us, not gcc. If you
compile
the same project on the same machine with the same gcc, using Linux and
QNX6, you'll see significant differences. So there are some bottlenecks
in
things like IO than need to be improved on our end as well.

That's encouraging, as it means to me you can do something about it ;-)
We're always looking. The OS is very stable now so much more time can be
devoted to performance tweaking. There are features inherent to a realtime
message passing architecture that will tend to make it slower for some
applications but I believe we are closing the gap. I've been using the same
machine for RTP for two years now and have seen things improve in many ways.
Recently some of our guys were asked to try and optimize the performance of
Eclipse on RTP and in very short order were able to make it comparable to
Windows and Solaris. In short, performance is as important to us as it is
to you but stability and reliability have to come first.

Kris

Thank you very much for your reply anyway...

cheers

Joern








Kris Warkentin

Re: gcc-3.1

Post by Kris Warkentin » Tue May 28, 2002 6:51 pm

"Rennie Allen" <rallen@csical.com> wrote in message
news:3CF3530F.3060205@csical.com...
Kris Warkentin wrote:

Recently some of our guys were asked to try and optimize the performance
of
Eclipse on RTP and in very short order were able to make it comparable
to
Windows and Solaris.

Wow ! This is good news. Both in that mgmt at QSSL still views
self-hosted as enough of a priority to issue such an edict, and that I
am currently in the process of downloading it. Now if only we could
optimize the internet such that downloading a 640MB image didn't take so
long :-)
Well, don't forget, self-hosted is just QNX6 on x86, so any optimizations
that are made in the OS at large also apply to self hosted. If you feel you
need more encouragement, consider the following two points: We've gone to
the trouble of making the Eclipse IDE available on Neutrino. We've done all
our packaging for the Windows and Solaris SDK in the format that is used by
the QNX6 packager.

I think self hosted is not only good for external developers but, by 'eating
our own dogfood' in house, we also improve the quality of the product. I
recall several years ago as more and more people started using RTP as their
development host in house, the rate of improvement increased exponentially
because so many eyes were on it.

cheers,

Kris
Rennie

Igor Kovalenko

Re: gcc-3.1

Post by Igor Kovalenko » Fri May 31, 2002 12:56 am

"Kris Warkentin" <kewarken@qnx.com> wrote in message
news:ad0409$63c$1@nntp.qnx.com...
"Mario Charest" <goto@nothingness.com> wrote in message
news:ad02pq$ec5$1@inn.qnx.com...

That's encouraging, as it means to me you can do something about it ;-)

We're always looking. The OS is very stable now so much more time can be
devoted to performance tweaking. There are features inherent to a
realtime
message passing architecture that will tend to make it slower for some
applications but I believe we are closing the gap.
There are some inherent things indeed, but I doubt they contribute more than
10% to the actual slowness. Higher rate of context switches due to message
passing tends to hurt cache locality for example. However, most of major
slowdowns occurs elsewhere. To name a few:

- you have relatively slow malloc() and even slower realloc();
- you have miserable memmove();
- block I/O layer is very inefficient (causes too many short messages to be
passed, apparently);
- video drivers do not use all acceleration available in hardware (I don't
think they really use AGP either, do they?);

There is nothing inherent in that. Fix it and speed difference with Linux
will be hardly noticeable, especially since Linux code tends to do stupid
things which I suspect would still make it slower than properly designed
message-passing OS.

-- igor

Jörn Ihlenburg

Re: gcc-3.1

Post by Jörn Ihlenburg » Fri May 31, 2002 11:37 am

Hi Kris,

"Kris Warkentin" <kewarken@qnx.com> schrieb im Newsbeitrag
news:ad010v$3mk$1@nntp.qnx.com...
The dream may be closer than you think - our binutils patch is currently
in
FSF hands, hopefully just needing some tweaking (fingers crossed ;-)
before
they accept it into their main branch.
I have been playing with GCJ on Linux and it's really impressive. An articel
from IBM about building Apps with Java and SWT and compiling them with GCJ
into native binarys got me on the way. Under Linux it's actually faster
(with -O2) than the JIT-Compiler of the VM from Sun (which is usually VERY
fast). I hav to do heavy byte array processing and GCJ is double as fast as
Sun's VM.
If I could get GCJ running under QNX, i would suspect some performance
improvements over IBM's J9 as well. Another good thing is that memory
footprint is dramatically reduced. If I use J9 under QNX with JIT-Compiler
activated, it eats up a LOT of memory... And memory is always precious :)
I'll look forward to test it under QNX. Many thanks for your instant
relpy...

Joern

Post Reply

Return to “qdn.public.qnxrtp.porting”