Re: New redpaper: i5/OS Program Conversion: Getting ready for i5/ OSV6R1

Can you even fathom vendors compiling multiple versions of their software
- V6R1 non power 6 and V6R1 Power 6 (Let alone compiling down to V5R3 or
earlier)?

See also last couple of sentences at the bottom about "under used"
feature.


From the "raw" redpiece (thus highly subject to change).


If you compile your application on a POWER6-based system, it may run
faster on such
systems than it otherwise would. This depends on how extensively your
application ends up
using hardware features that are found only on POWER6 processors. However,
it must then
be converted if it is loaded on a system with an older processor. So, your
installation
instructions should discuss the difference in load times, if that is
significant for the programs
in your application.
If, for example, your application does not improve much in performance
when using
POWER6-specific features, then it may be best to create it such that it
will use only features
available on all supported processors. That approach will avoid conversion
time when the
programs are loaded on non-POWER6 systems running V6R1. You can either
create the
programs on a system with an older processor, or “opt out” of ACG when
creating programs
on a POWER6 processor. The latter can be done by setting the
QIBM_BN_CREATE_WITH_COMMON_CODEGEN environment variable to a value of 2,
or
by specifying the LICOPT(‘CodeGenTarget=Common’) parameter on the CRT* or
CHG*
commands you use when building your application.
The new create options might significantly improve the performance of your
application.
However, keep in mind that they are only available for programs that
target V6R1.
You may also want to consider application profiling as a way to improve
performance. While it
is not new, it is an under used feature that has resulted in overall
application performance
improvements up to 10%. See the ILE Concepts manual chapter on Advanced
Optimizations
for more information.

Rob Berendt