On Tue, Oct 16, 2012 at 10:19:54AM -0700, Ian Lance Taylor wrote:
> On Mon, Oct 15, 2012 at 5:28 AM, Aldy Hernandez <al...@redhat.com> wrote:
> I don't see a reason to add this new function. I would just inline it
> into noce_can_store_speculate_p, replacing the call to
> memory_modified_in_p. And I'm not sure I see a reason to change the
> comment for memory_modified_in_p, it seems to already be accurate.
>
> Clearly we could consider the possibility of a PARALLEL of SET insns,
> but of course most the compiler won't handle that anyhow. I suppose
> that would be a reason to use memory_surely_modified_in_insn_p, but in
> that case you might as well handle the PARALLEL case now.
I see another problem with noce_can_store_speculate_p, the return false;
checks. The search for a store to exactly that location can be done
just on post dominator bbs, but I wonder how you can avoid looking for
volatile insns or non-const/pure calls (or insns that may modify
the address) also in other bbs.
What if there is:
location of considered if-conversion on mem
if (some_condition)
call ();
mem = something;
? We only walk immediate post-dominators, so look at the bb containing
mem = something;, but I think a call/volatile insn/modification of mem's
address is problematic in any bb on any path in between test_bb and
the post-dominator on which the write has been found.
Jakub
