------- Comment #4 from zero at colonel-panic dot org 2007-10-30 16:45 ------- Subject: Re: gcc-4.2.2 generates bad code on ARM
rguenth at gcc dot gnu dot org wrote: > ------- Comment #3 from rguenth at gcc dot gnu dot org 2007-10-30 15:54 > ------- > It doesn't matter that you store the address of param to a volatile variable > (this store is not removed), but that > > DMA_ADDR_REG = (unsigned long) data; > > does not make the stack local live longer than the frame of outer_func and > *data is not used after the assignment. Probably your code kicking off > the DMA engine lacks a memory barrier and a use of DMA_ADDR_REG. > > Thanks for the explanation. Seems like an optimisation too far. I can't see that using the address of a variable and not it's contents is a particularly common practice, and having to use a GCC specific barrier to get simple embedded code to work sucks. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33949
