Hi Bernd, You may send a patch too.Your idea will be more clear. -- Regards lin zuojian
On Sun, Mar 02, 2014 at 10:24:52AM +0800, lin zuojian wrote: > Hi Bernd, > > set_mem_align is not working like magic. > > set_mem_align just set the alignment of a memory rtx.And You must aware > that you do so because you are sure this rtx IS aligned like this. > For arm machines, the base the virtual registers are aligned to 8 > bytes.You can't just set_mem_align to get_best_mode(),because this is > not the fact.If you do so,the unalign accessing is still there.My > realignment code make sure the base of virtual registers is aligned to > SImode,and so I can use that fact to set_mem_align. > > In a word,set_mem_align does not generate realignment code for > STRICT_ALIGNMENT machines. > > As you say that's not the best mode on all machines.But let's fix one > bug at a time.You are focusing too much in that mode. > > > So, that is what I mean: this patch makes the stack grow by > > 32 bytes, just because the emit_stack_protection uses SImode, > > with unaligned addresses which is not possible for ARM, and > > not optimal for X86_64. > It's because of the realignment,as my comment said.I have to make room > for virtual registers's realignment,or they(or the spilled registers) will > get damaged by the other subroutine. > > > -- > Regards > lin zuojian
