On 06/06/2011 10:15 AM, Andreas Färber wrote: > Am 06.06.2011 um 16:25 schrieb Paolo Bonzini: >> +#if defined(__arm__) || defined(__sparc__) || defined(__mips__) || >> defined(__hppa__) || defined(__ia64__) >> +#define WORDS_ALIGNED >> +#endif > > Since it depends on the host and you're placing it directly under > config-host.h inclusion, might it make sense to move the decision > into configure instead, so that it ends up in config-host.h?
Hum, I now understand what Paulo was talking about elsewhere in the thread. If he takes my suggestion to re-write the unaligned functions with GCC packed support, these host ifdefs go away, and this objection disappears. The question becomes one of ordering. Do we take his existing 3-part patch as-is, and the packed patch as a followup? Do we convert to packed accesses first and move it around after? Do we do it all in one step? r~
