#265 ReSID cpu/arch configure problem

[André Steinert]

I've got a problem with ReSID if I compile VICE 2.3.12 on an Intel Core i7 using MingW32 under Windows 7 64-bit. On the computer I did the compile, VICE runs flawlessly. If I try to start VICE on an Pentium M having ReSID active, the program quits with "An unexpected error occured. Received signal 4 ().", but it works if I change the SID to ReSID-fp or FastSID. Another clue indicating ReSID being the problem could be, that nearly all object files of ReSID are different from the ones of a compile on the Pentium M, whereas many of the object files of ReSID-fp and FastSID are identical. It works the other way around: If I compile VICE on the Pentium M, it works (even with ReSID) on the Core i7.

[gpz]

not sure if that is the problem, but resid is (afaik) the only part in vice that uses SSE stuff - maybe if you compile on 64bit host gcc uses 64bit instructions that simply dont work on 32bit systems.

[Anonymous]

> not sure if that is the problem but resid is (afaik) the only part in vice that uses SSE stuff

Wrong, but on the right track. reSID-fp is the only thing that uses SSE explicitly and that's only conditionally after cpuid checks. The problem is defaulting to -march=native which triggers a bunch of CPU specific optimizations, notably affecting reSID. Use "./configure --enable-arch=no" to disable or try something like "--enable-arch=pentium3" for a common base.

As a sidenote, this is not a win32-only problem. See the Debian bug tracker for a giggle. I suppose "use Gentoo" is not an acceptable workaround.

[gpz]

that sounds reasonable - so the autocrap magic needs some fixing i guess =) -> assigning to blacky then

[Marco van den Heuvel]

Changed the default of --enable-arch to be 'no' in rev 26058.