[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: failure of cross-compilation detection on BlueGene/L

From: Steven G. Johnson
Subject: Re: failure of cross-compilation detection on BlueGene/L
Date: Sat, 06 Jun 2009 11:42:23 -0400
User-agent: Thunderbird (Macintosh/20090302)

Christian Rössel wrote:
Therefore I usually specify --host *and* --build if I want to cross compile. If $host != $build, configure sets cross_compiling to yes

Ah yes, that is an easier workaround than hacking the configure script.

However, this doesn't change the fact that there is a bug: autoconf's detection of cross-compiling when cross_compiling=maybe should be fixed on BlueGene. Specifying --build is only a workaround, not a fix.

Note that we are working on some heuristics to detect supercomputer platforms like Blue Gene, Cray, NEC etc. that require cross compilation. This detection will be available as a macro soon and render it unnecessary that the user specifies --host and --build on the supported platforms.

I'm not sure that's the right approach. The default when running configure should be to compile for the build host (i.e., the front-end nodes on BlueGene etc.). Also, the autoconf philosophy is to do feature tests rather than platform-specific heuristics whenever possible.

How do the autoconf developers feel about my proposed solution, which should allow --host to work normally on BlueGene?


reply via email to

[Prev in Thread] Current Thread [Next in Thread]