------------------------------<snip>-------------------------
I have problems with this assertion.
Reentrant code tends to be larger than non-reentrant code and to run slower
in that it generally has to increase the instruction path length due to the
need to acquire and release dynamic storage for work areas and to copy model
data into that dynamic storage.
It is true that you can share copies of a reentrant module across address
spaces, which can eliminate the need for repeated LOAD requests, and can
save memory.
However, an evaluation of how and where the module is to be used is needed
in order to determine whether writing the module to be reentrant is
justifiable.
A flat assertion of better performance is not supportable.
--------------------------<unsnip>--------------------------------
John, it's a function of the internal cache usage; z900 and later
processors have a separate caches for instruction fetch and data fetch;
usage DOES have a significant effect on performance. While the factors
you cite can have an effect, experimentation has born out the fact that
the basic assertion is valid.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
