RE: RISC instruction stream hash value -- MI400

Hi Will,

on the issue of speed, you are not going to get the best results in MI vs an
ILE/wcode language.  This is because of the instruction piping/parellel
processing of the power pc risc chip.  MI code cannot be optimized like ILE
code due to the way exceptions are handled in MI vs ILE.  Think of what
happens when instructions are executing in parellel and a divide by zero
error occurs in one of the instructions.  In MI, exception handler code is
called and basically the program continues on its way depending on what the
excp handling code does.  But if instructions were running in parellel, that
is instructions /0001 thru /0004 were running at the same time, when insn
/0002 bombs the excp handler expects that instructions /0003 and /0004 have
not been executed.  But they might have been, so there are going to be
problems in the program.

In an ILE stack based pgm model this guarantee-ing of instruction exec
sequence is not an issue when the exception is thrown thrown from one
function call level back to its caller.  All the results of the parellel
instructions in the function that bombed are on the stack ( automatic
storage variables of the function ) and throwing the exception back to the
caller pops that stack and makes those results disappear.

Now if MI variables could be coded as "mutable" or something to indicate to
the optimizer that a variable does not need its value guaranteed at
exception time ( the MI program implements its own stack ), then the
optimizer could optimize those instructions that have such a variable as an
lvalue ( left, update side of the expression ).

If you are patching the code at the power pc/ risc level then that sounds
pretty interesting just as an academic exercise. Some people on this list
have done that, but I dont know much myself.  You should look thru the MI
mailing list archives.

Something else to consider is to get the PowerPc assembler compiler that
runs on the other power pc lines of computers that ibm sells.  I dont know
if such a thing is obtainable, but if it is it would be interesting to see
if its code could be made to run on the iSeries.

Good luck and post as much of what you find out as you can.

Steve Richter


-----Original Message-----
From: mi400-admin@midrange.com [mailto:mi400-admin@midrange.com]On
Behalf Of Will Richards
Sent: Thursday, December 19, 2002 10:31 AM
To: mi400@midrange.com
Subject: [MI400] RISC instruction stream hash value


Hi Kurt,

Thanks for your reply.

I'm writing a MI program which is suppose to replace an old and very slow
C/400 that solves an ordinary differential equation. MI program works fine
and A LOT MORE FASTER that C, :-). Looking for ways to improve speed and
being familiar with RISC, I've patched the MI program and made it even
faster. Now, the problem is how to deploy it to all our systems without
being flagged as altered by the restore operation. My guess is, and please
correct me if I'm wrong, if we recalculate the RISC instructions hash
(Procedure table, offset X'009C' ) everything will be OK.

Thanks again,

Will







--
WillDRichards@netscape.net


__________________________________________________________________
The NEW Netscape 7.0 browser is now available. Upgrade now!
http://channels.netscape.com/ns/browsers/download.jsp

Get your own FREE, personal Netscape Mail account today at
http://webmail.netscape.com/
_______________________________________________
This is the MI Programming on the AS400 / iSeries (MI400) mailing list
To post a message email: MI400@midrange.com
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/cgi-bin/listinfo/mi400
or email: MI400-request@midrange.com
Before posting, please take a moment to review the archives
at http://archive.midrange.com/mi400.