Re: RFA: New optimization: Convert local statics into non-statics

[Ian Lance Taylor <iant at google dot com>]

Robert Dewar <dewar@adacore.com> writes:

> I think an entire separate pass that is specifically benchmark oriented
> seems a bad idea, unless there are real world programs, I would forget
> about this optimization, unless it falls out of some more general
> optimization.

gcc vendors focused on the embedded space are going to care about
their performance on the EEMBC benchmark.  If we don't put this
optimization into generic gcc, they are going to maintain separate
versions of it.  This doesn't seem particularly useful either.

Clearly there is a cost-benefit tradeoff here.  But I don't think we
should rule out benchmark oriented optimizations entirely.  I think we
should consider the low benefit against the cost.

For what it's worth, this benchmark was a real world program at one
time, although it's hard to understand why anybody would write code
that way.

Ian