Re: RFC/RFA: patch for PR 22156: improve SRA for bit-fields

[Alexandre Oliva <aoliva at redhat dot com>]

On May  6, 2007, Eric Botcazou <ebotcazou@adacore.com> wrote:

>> No, you talked about ICEs.  I regularly build the Ada compiler (and I
>> did even at the time it was broken in mainline because I was working
>> on a tree from before the breakage), and I haven't seen any such ICEs.

> I didn't talk about ICEs first, only about pessimization.  And I gave you a 
> testcase and explained what happened.  Why didn't you simply look at it?

Because the symptoms you were describing looked perfectly fine to me.
I assumed I'd just see what you were telling me I'd see, so why would
I spend time duplicating what I already knew?

>> The BIT_FIELD_REFs you're talking about are a separate issue, that
>> follows from a misunderstanding of the purpose of the patch.

> I don't think so: your patch does create BIT_FIELD_REFs out of
> nowhere in situation where it blatantly shouldn't,

It's the "out of nowhere" that we disagree.  You seem to assume that
BIT_FIELD_REF is only to be used for bitfields, but that's an
incorrect assumption.

> In the testcase, the structure contains exactly 2 fields, which are both 
> words.  How does this relate to what you're saying?

Two SImode fields that are laid out in such a way that they can be
accessed with a single DImode access.  The first patch used to
consider it profitable to create a single variable for them.

With the latest two patchlets on top of the big patch I last posted,
it won't, and will rather avoid creating blocks wider than a word.
This should appease your concerns.  But it will still create a single
variable for say a sequence of 4 chars (assuming a word is 4 bytes
long).

> All right, but let's do it carefully then.

Yes, sir ;-)  been trying to, honest ;-)

>> But how bad do we do after the patch, and how does it compare with
>> what we did before the patch?

> Which patch are you talking about exactly?  You have posted quite a few as of 
> this writing. :-)

I'll post a consolidated patch momentarily, upon Diego's request.

-- 
Alexandre Oliva         http://www.lsd.ic.unicamp.br/~oliva/
FSF Latin America Board Member         http://www.fsfla.org/
Red Hat Compiler Engineer   aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist  oliva@{lsd.ic.unicamp.br, gnu.org}