Re: [MI400] optimizing MI code

["Paul Godtland" <plg@xxxxxxxxxx>]

Leif Svalgaard asked:

|couldn't you use a OVRPGATR-like mechanism to turn on/off
|optimization? The programmer could "promise" with a OVRPGATR
|that no exceptions will occur from now on (or that he doesn't care
|about things being correct if an exception *does* occur - like
|turning of binary size checking). This will allow optimization to
|proceed at the programmer's peril.

There are two reasons why I don't think function like you suggest will be
added. First, on our single-level-store system, a choice about potential
addressing problems cannot safely be left to individual programmers,
because others could also be affected. Second, it would take significant
development resource to make sure we got even the non-address data
optimization completely correct, and that just to improve one aspect of
optimization for original MI. Fully optimizable alternatives are already
provided. For example, ILE C supports direct use of MI built-ins, too (and
then  header files are usually provided for declaring the templates). So,
our limited development resource is better applied to other work that will
benefit more customers.

I poured a lot of time, energy, and at times, high levels of stress into
address optimization for original MI on IMPI machines, so it was near and
dear to me. There were always more improvements I wanted to make, even as
we designed ILE, but I got over that. ;-) So, I'm sure other long-time MI
programmers will also come to accept its optimization limitations and move
to the better alternatives available for situations that require excellent
optimization.

Paul Godtland