How to test if preprocessor defines foobar in macro?

[Dr. David Kirkby]

I'm trying to modify this macro

http://git.savannah.gnu.org/gitweb/?p=autoconf-archive.git;a=blob_plain;f=m4/ax_count_cpus.m4

which attempt to get the number of CPUs in a system. I'd like to extend
it to cover Solaris, AIX and HP-UX.

It seems sensible to me to only do a test on a platform it will work on,
so I think thinking of something like

if `uname` = "HP-UX"
   write test for HP-UX
fi


or I could check if __hpux__ is defined or not.

Can anyone show me how to modify the above autoconf macro to do the test
using either uname, a macro that the pre-processor defines, or other
that allows me to execute a bit of code on only one platform. method.


There's another macro I've seen used for determine cache sizes of
processors.

http://www.nongnu.org/autoconf-archive/ax_cache_size.html

which depends on the CPU being an x86 one, which it will often not be. I
don't fancy the task of trying to work out cache sizes for every
processor used on machines running Solaris, AIX or HP-UX, so will try to
modify that so it returns a sensible guess (say 1 MB) on processors it
does not know about. The test is causing a script to crash on HP-UX.

Dave


_______________________________________________
Autoconf mailing list
Autoconf@...
http://lists.gnu.org/mailman/listinfo/autoconf

[Ralf Wildenhues]

* Dr. David Kirkby wrote on Wed, Oct 14, 2009 at 06:36:36AM CEST:
> http://git.savannah.gnu.org/gitweb/?p=autoconf-archive.git;a=blob_plain;f=m4/ax_count_cpus.m4

This macro does not look well-designed.  It does the wrong thing when
cross-compiling, or even only compiling for a different system (which
happens to be use the same architecture).

A better approach would be to provide the user with a switch, or prefer
a user override of CPU_COUNT.  If that switch/variable was not used, and
cross compilation is not enabled, then one can try to use the $build
system features; but even then you should be cautious about the result.

> which attempt to get the number of CPUs in a system. I'd like to
> extend it to cover Solaris, AIX and HP-UX.

How do you find out this information on those systems?

> It seems sensible to me to only do a test on a platform it will work
> on, so I think thinking of something like
>
> if `uname` = "HP-UX"
>   write test for HP-UX
> fi

uname gives you $build system info.  AC_CANONICAL_HOST sets $host which
helps better, and gives fairly uniform info.  See config.guess for a
number of strings to match.

> There's another macro I've seen used for determine cache sizes of
> processors.
>
> http://www.nongnu.org/autoconf-archive/ax_cache_size.html

This has similar problems.

> which depends on the CPU being an x86 one,

Not only that.  IIUC it also depends on running on non-ancient Linux
only.

> which it will often not
> be. I don't fancy the task of trying to work out cache sizes for
> every processor used on machines running Solaris, AIX or HP-UX, so
> will try to modify that so it returns a sensible guess (say 1 MB) on
> processors it does not know about. The test is causing a script to
> crash on HP-UX.

HTH.

Cheers,
Ralf


_______________________________________________
Autoconf mailing list
Autoconf@...
http://lists.gnu.org/mailman/listinfo/autoconf

[Noah Misch-2]

On Wed, Oct 14, 2009 at 05:36:36AM +0100, Dr. David Kirkby wrote:

> I'm trying to modify this macro
>
> http://git.savannah.gnu.org/gitweb/?p=autoconf-archive.git;a=blob_plain;f=m4/ax_count_cpus.m4
>
> which attempt to get the number of CPUs in a system. I'd like to extend  
> it to cover Solaris, AIX and HP-UX.
>
> It seems sensible to me to only do a test on a platform it will work on,  
> so I think thinking of something like
>
> if `uname` = "HP-UX"
>   write test for HP-UX
> fi
>
> or I could check if __hpux__ is defined or not.

See AC_CANONICAL_HOST for the way of identifying the host platform.  However, we
generally omit explicit assumptions about the right test for a certain system
type.  Instead, try each test unconditionally and use the first one that works.
The code you reference currently does this.

Consider checking processor count and cache characteristics at runtime.  If a OS
distribution ships binaries of your software, they will run on a wide range of
processor configurations.  You might need `configure'-time checks to verify the
availability of the APIs you would use at runtime.


_______________________________________________
Autoconf mailing list
Autoconf@...
http://lists.gnu.org/mailman/listinfo/autoconf

[Mike Frysinger]

On Wednesday 14 October 2009 00:36:36 Dr. David Kirkby wrote:
> I'm trying to modify this macro
>
> http://git.savannah.gnu.org/gitweb/?p=autoconf-archive.git;a=blob_plain;f=m
> 4/ax_count_cpus.m4
>
> which attempt to get the number of CPUs in a system. I'd like to extend
> it to cover Solaris, AIX and HP-UX.

there are so many reasons why this is wrong (some of which people have pointed
out).  you havent really said why you want this information, but presumably it
isnt for usage in the configure script (since that's about the only place this
info would be valid).  if it's being passed to source code, you should be
querying the proc count at runtime with sysconf(_SC_NPROCESSORS_CONF).  if
your system isnt POSIX complaint, then usually it provides some other way of
getting this info.
-mike


_______________________________________________
Autoconf mailing list
Autoconf@...
http://lists.gnu.org/mailman/listinfo/autoconf