This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] combine: Special handling for conditional refs when creating log links
On 05/07/2010 07:33 PM, Steven Bosscher wrote:
> On Fri, May 7, 2010 at 7:13 PM, Bernd Schmidt <bernds@codesourcery.com> wrote:
>> Fact, actually. We did have code in flow.c to track liveness in the
>> presence of COND_EXEC exactly, and it appears to have been lost entirely
>> when df was introduced.
>
> It depends on what you mean with exactly. I suppose you mean per insn,
> like in df_simulate_one_insn_backwards. That was apparently lost, yes.
> I don't know why.
Insufficient patch review?
Exactly means that if you have, for example,
(use reg A)
(lots of stuff)
((cond NE) (set (reg A) (something)))
((cond EQ) (set (reg A) (something)))
it knows that register A is dead after the first use. I don't see
anything in df that would keep track of this.
>> I've recently discovered this when looking at
>> peephole2; see my recent mail about this.
>
> I assumed you are referring to
> http://gcc.gnu.org/ml/gcc-patches/2010-04/msg01395.html.
Yes.
Bernd