i686-apple-darwin9 bootstrap is broken since r162678 (with an ICE reported for other targets as well). Since this was fixed (at r162697), it has unmasked another fail which is still present at 162778. unfortunately, the other bootstrap bug prevents narrowing this down more... however; This is a compare-debug fail of cfgexpand.o. Looking at the .s files (to be attached) there seem to be instruction ordering differences. this bug is _not_ present in powerpc-apple-darwin9 or x86_64-apple-darwin10.
Created attachment 21364 [details] .i and .s files from cfgexpand for stage2 & 3
cc-ing Bernd having reviewed the svn logs between r162678 and 162697.
this is a bit tedious to work through ... but for example in _expand_one_stack_var_at: we have for stage2 : ************* cmovae %edx, %eax # tmp143,, max_align cmpl $0, 52(%esp) #, %sfp jbe L252 #, L245: movl %eax, 48(%esp) # max_align, %sfp L227: movzwl %cx, %eax # D.53575, tmp154 sall $6, %eax #, tmp154 addl L_tree_contains_struct$non_lazy_ptr-L00000000044$pb(%ebx), %eax #, tmp155 *********** and for stage 3: cmovae %edx, %eax # tmp143,, max_align LM516: cmpl $0, 52(%esp) #, %sfp jbe L252 #, L245: LM517: movl %eax, 48(%esp) # max_align, %sfp movl $0, 52(%esp) #, %sfp L227: LVL348: LBB2461: LM518: movzwl %cx, %eax # D.53575, tmp154 sall $6, %eax #, tmp154 addl L_tree_contains_struct$non_lazy_ptr-L00000000044$pb(%ebx), %eax #, tmp155
I back-ported r162697 to r162678 and see comparison fail at r162678.
*** This bug has been marked as a duplicate of 45162 ***