CC's very recent post was timely.
I spent three days recently bailing out a dubious conversion of some not very
good TCB-driven code to run under just one SRB.
The converter had not understood much of anything about what he was doing; and
his problems were aggravated by the fact that he did not understand that
floating-point registers must be saved when they are heavily used even if that
use is implicit, in this case for some pie-chart arcsins and actuarial
exponentials.
In the end I had to rewrite the "scientific" code (using standard
Chebychev-Hastings approximations) because the LE routines being used for these
functions had not, of course, been designed to run under an SRB. I hope it
will be a very long time before I encounter another requirement to evaluate
exponentials under an SRB, but I now have the routines in hand should this
dubious requirement arise again.
This experience left me convinced that there will be a large new market for
services that do such conversions competently and knowledgeably. Meaningful
savings are achievable, but not by the unwashed.
Great patience with the clients who need these services will be required.
Among other things, these clients will perforce have to come to understand that
radical redesign and the use of many SRB-driven routines to replace single
TCB-driven ones are both necessary.
It should be, but clearly is not, obvious that the simplistic notion of
one-for-one conversion is a recipe for disaster.
John Gilmore Ashland, MA 01721-1817 USA
_________________________________________________________________
The New Busy is not the old busy. Search, chat and e-mail from your inbox.
http://www.windowslive.com/campaign/thenewbusy?ocid=PID28326::T:WLMTAGL:ON:WL:en-US:WM_HMP:042010_3
----------------------------------------------------------------------
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
