RE: [MI400] optimizing MI code

["Steve Richter" <srichter@xxxxxxxxxxxxx>]

-----Original Message-----
From: mi400-admin@midrange.com [mailto:mi400-admin@midrange.com]On
Behalf Of Paul Godtland

>W-Code is MUCH more optimizable, for at least two key reasons. First, it's
>expression-based instead of storage-based. That is, results are viewed as
>being returned to a logical (not necessarily physical) stack, not to the
>storage location specified as an operand on the instruction.

Does that mean the results can be retained in the registers of the cpu ?

>Secondly, the
>exception handling semantics are very different. One example: in W-Code
>there is no concept of handling an exception by transferring control to the
>"next" MI (or W-Code) instruction.

So exceptions are only handled on scope boundaries ?  Is that where
lumpiness comes in ?

This matters because the optimized code may execute more than 1 stmt at a
time or out of sequence and if an exception occurs, all the out of sequence
and parallel instruction changes have to be rolled back to as of the
instruction that caused the exception. If all the variables are on the scope
stack, the changes can all be undone by poping the stack and throwing the
exception back to the scope caller.

Is that correct ?  And if so, are pointer based variables not optimizable ?
So that in the following code:

void ChangeNumbers( int* pnNumber1, int* pnNumber2, int* pnNumber3 )
  {
  *pnNumber1 = 5 ;
  *pnNumber2 = 10 ;
  *pnNumber3 = 15 ;
  }

in case of an exception on the 2nd stmt ( pointer not set ), stmt1 must have
completed execution and stmt 3 must not be executed.

thank you Paul,

Steve Richter