[LLVMdev] alloc_size metadata

[Nuno Lopes]

>> About Duncan's comment about having the memory builtin analysis
>> recognize this intrinsic, well I agree it should (and I'll take care
>> of that), but I'm not sure if we should be very aggressive in
>> optimizing based on this metadata.
>> For example, do we really want to remove a call to a custom
>> allocator whose return value is unused (like we do for malloc)?  If
>> so, we'll also need a metadata node to mark de-allocation functions
>> (so that sequences like my_free(my_malloc(xx)) are removed).
>
> You might want to restrict that to custom allocators that have an
> additional 'removable' attribute.
>
> Out of curiosity, does any of this potentially effect the presence or
> absence of 'inbounds' on GEP instructions?

No, I don't think so.  The inbounds flag means that the offset computation 
will not overflow. Here we are only discussing the discovery of the object 
size (potentially pointed by a GEP). As Duncan pointed out, knowing an 
object's size may be used to expose more undefined behavior to optimizers if 
we prove that the index will fall out of bounds.

Nuno