It seems that gst-plugins-base 1.0.3 is actually built, not 1.0.5. (Not that I think it will make a difference for this particular issue, just saying).

> {standard input}: Assembler messages:
> {standard input}:655: Error: immediate value out of range -- `movt ip,-32640'
> make[5]: *** [libgstvideo_1.0_la-tmp-orc.lo] Error 1

This happens whilst compiling a plain .c file, so I would say this is a toolchain/compiler bug.

You're right, the specific logfile is from 1.0.3, I retried with 1.0.5, got the same error and updated the bug's title w/o uploading a new logfile.
What I think is strange:
the line just below the error states: make[5]: *** [libgstvideo_1.0_la-tmp-orc.lo] Error 1
so it looks to me like orc-optimization is enabled while "--disable-orc" is specified on the configure command line (and the build log states this again: "Orc acceleration disabled by --disable-orc.  Slower code paths will be used.")

I honestly don't know anything about orc, what it is and does but might it be it only works on x86/x86_64 and that's why the assembler (not the compiler) throws up here?

unfortunately the arm-box is powered off at home now so I can't directly work on it

liborc generates machine code at runtime. What is being compiled here are the backup functions, which are plain C, and get used when liborc is not available or disabled, or when liborc couldn't generate code at runtime for a certain orc function (e.g. because there aren't enough registers, or an opcode isn't mapped for the specific architecture).

Indeed, this looks like a bug in gcc. A bit of searching around on the net confirms this. Maybe it's time I install gcc-4.6 on that machine. Thanks for the help anyways :)