Re: PR 23551: why should we coalesce inlined variables?

[Alexandre Oliva <aoliva at redhat dot com>]

On Jul  8, 2007, "Andrew Pinski" <pinskia@gmail.com> wrote:

> I really wish you added a testcase for the patch which you were
> working on because that would have shown why we would want this patch

Actually, IMHO the burden of proof was on why we would want to
distinguish between inlined variables and non-inlined variables.
When that distinction was introduced, it made a significant difference
for performance.  Now, it no longer does.  So the reason to have that
artificial distinction was gone.  So we're back on why should we want
to make this distinction.

And if the answer is because we want to privilege reduction of memory
consumption over the ability to generate reasonable debug info, then
we might as well pursue that and remove the entire distiction on user
variables, since it's there only to aid debug info.

And if we want to find some spot in the middle, we'd better have a
good explanation for both those who'd rather have better performance,
lower memory use or better debug info.

> and you said yourself does not help debugging info at all

Not quite.  I just couldn't find a simple testcase in which it made a
difference.  Mostly, it didn't make any difference, but even in the
performance analysis I posted I showed there were some (small)
differences.  Surely not worth a 90% memory increase, but has anyone
actually compared the debug info of this testcase in both situations
before overgeneralizing both the increase in memory use and the
absence of effect in debug information?

> so if it does not help then how do we make progress here?

Keep removing other obstacles to the generation of good debug
information.  Do you honestly expect it to be a one-shot patch?

-- 
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}