gnat cross-compiles ?

gnat cross-compiles ?

[Marc Espie]

On what arches does gnat currently work ?
I'd like to try to build it on openbsd architectures, I have i386
binaries... I've tried cross-building i386->sparc64, with little success.

I'd like to know which processors are happy with gnat in 3.1 (or the 3.1
branch), and whether cross-compiles have a chance to work, before I embark
on a wild goose chase...

Re: gnat cross-compiles ?

[Andreas Schwab]

Marc Espie <espie@nerim.net> writes:

|> I'd like to know which processors are happy with gnat in 3.1 (or the 3.1
|> branch), and whether cross-compiles have a chance to work, before I embark
|> on a wild goose chase...

We have already built gnat for ia64, alpha and ppc, via cross compiling
from i386.  Sparc, s390 and x86-64 are on my todo list.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: gnat cross-compiles ?

[Joe Buck]

Marc Espie <espie@nerim.net> writes:
> 
> |> I'd like to know which processors are happy with gnat in 3.1 (or the 3.1
> |> branch), and whether cross-compiles have a chance to work, before I embark
> |> on a wild goose chase...

Andreas Schwab writes:
> We have already built gnat for ia64, alpha and ppc, via cross compiling
> from i386.  Sparc, s390 and x86-64 are on my todo list.

I think that cross-compiling is the most confusing part of GCC, we
certainly get more questions about that than anything else, even from
experts.  Andreas, would you be willing to write a note describing your
process?  It doesn't need to be thorough, it could be as simple as just
listing the exact sequence of commands you used to produce, say, the ppc
compiler by cross-compilation from i386.

Re: gnat cross-compiles ?

[root]

Joe Buck wrote:
> 
> Marc Espie <espie@nerim.net> writes:
> >
> > |> I'd like to know which processors are happy with gnat in 3.1 (or the 3.1
> > |> branch), and whether cross-compiles have a chance to work, before I embark
> > |> on a wild goose chase...
> 
> Andreas Schwab writes:
> > We have already built gnat for ia64, alpha and ppc, via cross compiling
> > from i386.  Sparc, s390 and x86-64 are on my todo list.
> 
> I think that cross-compiling is the most confusing part of GCC, we
> certainly get more questions about that than anything else, even from
> experts.  Andreas, would you be willing to write a note describing your
> process?  It doesn't need to be thorough, it could be as simple as just
> listing the exact sequence of commands you used to produce, say, the ppc
> compiler by cross-compilation from i386.

I tried to build crosses for all of the RTEMS targets.  Here
is a post of my results.

http://gcc.gnu.org/ml/gcc/2002-05/msg00155.html

Summary: I get good results for sparc-rtems on ACATS.  I can build
a number of others.  Some have CPU specific failures (mips).  Other
CPUs do not have the exception processing support and fail in raise.c.
I would really like to know what is required to be supported to 
get by raise.c.  This is required to fix at least the i960, m68k and sh.
 
-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985

Re: gnat cross-compiles ?

[Andreas Schwab]

Joe Buck <Joe.Buck@synopsys.com> writes:

|> Marc Espie <espie@nerim.net> writes:
|> > 
|> > |> I'd like to know which processors are happy with gnat in 3.1 (or the 3.1
|> > |> branch), and whether cross-compiles have a chance to work, before I embark
|> > |> on a wild goose chase...
|> 
|> Andreas Schwab writes:
|> > We have already built gnat for ia64, alpha and ppc, via cross compiling
|> > from i386.  Sparc, s390 and x86-64 are on my todo list.
|> 
|> I think that cross-compiling is the most confusing part of GCC, we
|> certainly get more questions about that than anything else, even from
|> experts.  Andreas, would you be willing to write a note describing your
|> process?  It doesn't need to be thorough, it could be as simple as just
|> listing the exact sequence of commands you used to produce, say, the ppc
|> compiler by cross-compilation from i386.

I think the worst that can go wrong is that you *must* create ${libsubdir}
beforehand, otherwise the cross compiler does not find the target headers
(since all search paths are relative to ${libsubdir}).  Further, if you
don't use a unified binutils+gcc tree, you should create links
${builddir}/gcc/{nm,objdump} pointing to the respective target programs
before running configure, so that it can figure out the right
capabilities.  Everything else should be straight forward.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: gnat cross-compiles ?

[Geert Bosch]

On Thursday, June 6, 2002, at 09:46 , Marc Espie wrote:
> On what arches does gnat currently work ?
> I'd like to try to build it on openbsd architectures, I have i386
> binaries... I've tried cross-building i386->sparc64, with little 
> success.

There is a big difference between the compiler being able to bootstrap
and being a fully working port with proper support for all Ada features.
Especially in areas such as exception handling, stack-checking,
floating-point accuracy, tasking and interrupt support every system
requires special adaptation in the run time.

Even when systems claim support POSIX-compliant multi-threading for
example, there are always areas where support is insufficient,
either due to bugs, deviations from the standards or ambiguity
in POSIX specs. Finally, even if all features work on their own,
the combination may have problems due to lack of thread safety in
the underlying C run time or bugs related to signal handling in
multi-threading programs.

So, the answer to your question really depends on what you mean with
"works". Currently Sparc/Solaris (32-bit) and i386/Linux are in best
shape,  but even these are still not quite up to the level of GCC 2.8.1
based versions of GNAT. DEC Alpha, Sparc64, HP PA-RISC and PowerPC
based platforms should build.

   -Geert

Re: gnat cross-compiles ?

[Marc Espie]

On Fri, Jun 07, 2002 at 09:26:26AM -0400, Geert Bosch wrote:

> Even when systems claim support POSIX-compliant multi-threading for
> example, there are always areas where support is insufficient,
> either due to bugs, deviations from the standards or ambiguity
> in POSIX specs. Finally, even if all features work on their own,
> the combination may have problems due to lack of thread safety in
> the underlying C run time or bugs related to signal handling in
> multi-threading programs.

> So, the answer to your question really depends on what you mean with
> "works". Currently Sparc/Solaris (32-bit) and i386/Linux are in best
> shape,  but even these are still not quite up to the level of GCC 2.8.1
> based versions of GNAT. DEC Alpha, Sparc64, HP PA-RISC and PowerPC
> based platforms should build.

Anything like a testsuite, somewhere ?

Because, well, checking it beyond `it bootstraps' is a bit hard otherwise.

Re: gnat cross-compiles ?

[Joel Sherrill]

Geert Bosch wrote:
> 
> On Thursday, June 6, 2002, at 09:46 , Marc Espie wrote:
> > On what arches does gnat currently work ?
> > I'd like to try to build it on openbsd architectures, I have i386
> > binaries... I've tried cross-building i386->sparc64, with little
> > success.
> 
> There is a big difference between the compiler being able to bootstrap
> and being a fully working port with proper support for all Ada features.
> Especially in areas such as exception handling, stack-checking,
> floating-point accuracy, tasking and interrupt support every system
> requires special adaptation in the run time.
> 
> Even when systems claim support POSIX-compliant multi-threading for
> example, there are always areas where support is insufficient,
> either due to bugs, deviations from the standards or ambiguity
> in POSIX specs. Finally, even if all features work on their own,
> the combination may have problems due to lack of thread safety in
> the underlying C run time or bugs related to signal handling in
> multi-threading programs.
> 
> So, the answer to your question really depends on what you mean with
> "works". Currently Sparc/Solaris (32-bit) and i386/Linux are in best
> shape,  but even these are still not quite up to the level of GCC 2.8.1
> based versions of GNAT. DEC Alpha, Sparc64, HP PA-RISC and PowerPC
> based platforms should build.

Geert could you point me to something that explains what is
required to get past the missing exception handling support
that gets tripped on some targets with Ada enabled.

Thanks.

>    -Geert

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985

Re: gnat cross-compiles ?

[Rainer Orth]

dewar@gnat.com (Robert Dewar) writes:

> <<Anything like a testsuite, somewhere ?
> 
> Because, well, checking it beyond `it bootstraps' is a bit hard otherwise.
> >>
> 
> Yes, there is the ACATS suite. I believe that Laurent Guerby was
> adapting this for use in the gnu.org context?

Indeed: see

	ftp://ftp.oarcorp.com/pub/acats/acats4gnat-0.4.tgz

Right now, this is a standalone testsuite not yet integrated with DejaGNU.

	Rainer

Re: gnat cross-compiles ?

[Robert Dewar]

<<Anything like a testsuite, somewhere ?

Because, well, checking it beyond `it bootstraps' is a bit hard otherwise.
>>

Yes, there is the ACATS suite. I believe that Laurent Guerby was
adapting this for use in the gnu.org context?

Re: gnat cross-compiles ?

[Robert Dewar]

<<On what arches does gnat currently work ?
I'd like to try to build it on openbsd architectures, I have i386
binaries... I've tried cross-building i386->sparc64, with little success.
>>

Certainly should build on i386.
We have built the sparc64 compiler for Vxworks successfully also

Re: gnat cross-compiles ?

[Matthew Wilcox]

We have GNAT built on -debian-linux-gnu for:

  gnat-3.1 |    1:3.1-2 |      unstable | s390
  gnat-3.1 | 1:3.1.1-0pre1 |      unstable | alpha, hppa, i386, ia64, mips, mipsel, powerpc, sparc

arm & m68k both fail to build; i posted about it a while ago.

-- 
Revolutions do not require corporate support.