Question on TER

[Steven Bosscher]

On Thu, Sep 9, 2010 at 10:46 AM, Richard Guenther
<richard.guenther@gmail.com> wrote:
>> 1) Is this a valid optimization in general to attempt wrt other targets
>> (i.e. prevent replacement across calls)?
>
> I think it makes sense in general.

ISTR we even used to do so in the past...?

>> 2) Is there a way to recognize the __builtin_XXX() calls such that we know
>> they result in simple insns an not a real call?
>
> Unfortunately not easily.  What about just discounting all builtins?

Or allow replacing across a BUILT_IN_MD but not other built-ins.
BUILT_IN_NORMAL is more likely to result in a real function call than
an MD builtin.

I think you should allow TER across some GIMPLE calls if you can be
reasonably sure that the GIMPLE call will not result in an actual call
insn. It'd be a bit of "fake accuracy". For example, you also can't
avoid replacing across GIMPLE non-call statements that expand to
libcalls and such...

Ciao!
Steven