[MLton] experimental release 20051109

Matthew Fluet fluet@cs.cornell.edu
Tue, 15 Nov 2005 07:45:39 -0500 (EST)


>> Looks like a bug in should_duplicate_loop_header_p; it should not flag
>> volatile asm statements for duplication.  However, this is a corner
>> case because duplication of volatile asm statements usually does not
>> hurt, as long as only one duplicate is ever executed.
>>
>> I will check with the GCC folks.
>
> After looking at the situation myself and consulting with the GCC
> folks, I must say that MLton is at fault here.  GCC is permitted to
> duplicate asm statements. 8-(
>
> I'm not sure how to fix this, maybe using a special section for the
> profiling label addresses is the proper approach.

Could you say more about how a special section would address the problem? 
We need the profiling labels in the text section, since we want to map PC 
addresses back to profiling labels.