libmad ftbfs on armel in lucid

[Alexander Sack]

https://launchpad.net/ubuntu/+source/libmad/0.15.1b-4build1

due to bug 513734 ... and maybe others.

[Dave Martin (dave-martin-arm)]

Try this patch, which provides an alternative implementation of the 64-bit integer negation operation MAD_F_MLN which works around the lack of the RSC instruction in Thumb-2.

A missing earlyclobber constraing flag in the existing implementation is also fixed by the patch.

This fixes the build failure and I believe the code works... can anyone suggest how best to test it? (There seems to be no make check or make test).

[Dave Martin (dave-martin-arm)]

Oops, it looks like mad.h is a generated file... if so, the second part of the above patch should not be applied.

If mad.h is distrubuted in the source package, it should be rebuilt in the appropriate way from the other headers before uploading.

[Alexander Sack (asac)]

Dave, why this hunk below?

-# define FPM_INTEL
+# define FPM_ARM

this define isn't referenced anywhere else in the patch. Is this fixing a second issue?

[Dave Martin (dave-martin-arm)]

mad.h is a generated file and has # define FPM_<arch> depending on what the package was built for before the source was distributed... so it changed from INTEL to ARM after I did a build.

Since mad.h is generated from fixed.h during build, please throw away the second half of the patch (which modifies mad.h).

[Launchpad Janitor (janitor)]

This bug was fixed in the package libmad - 0.15.1b-4ubuntu1

---------------
libmad (0.15.1b-4ubuntu1) lucid; urgency=low

  * libmad.thumb.diff: use "adr" instead of "add" to make code ready for
    thumb2 (LP: #513734)
  * Provide-Thumb-2-alternative-code-for-MAD_F_MLN.diff: fix another
    ftbfs with thumb2 as "rsc" doesnt exist anymore - thanks to Dave
    Martin for this patch (LP: #534287).
 -- Alexander Sack <email address hidden> Fri, 19 Mar 2010 13:09:26 +0100