[Simh] Single stepping.

[Richard Cornwell]

Hi Mark,

> On Sunday, January 29, 2017 at 4:43 PM, Richard Cornwell wrote:
> > > On Sunday, January 29, 2017 at 12:50 PM, Richard Cornwell wrote:  
> > > >   How is single stepping implemented from the point of view of
> > > > the simulator?  
> > >
> > > Well, as long as you're managing decrements to sim_interval to
> > > reflect instructions executed, and you call sim_process_event
> > > when the sim_interval value is <= 0 and return from sim_instr()
> > > when sim_process_event () returns anything else but SCPE_OK, then
> > > it just works.
> > >
> > > Under the covers SCP has a fake unit which it schedules for the
> > > specified STEP count (default of 1) when you enter a STEP command
> > > at the sim> prompt.  That unit's service routine then gets
> > > dispatched to when the step count expires and it does nothing but
> > > return SCPE_STEP which should cause your above mentioned
> > > sim_process_event loop to return from sim_instr().
> > >
> > > Make sense?  
> > 
> >   That is what I thought. Now sim_interval for me counts memory
> > cycles, since instructions are variable length. However I have
> > noticed that "step" by itself does not always execute one
> > instruction. I wonder if maybe we should look at how single
> > stepping is implemented and perhaps pass an argument to sim_instr()
> > to indicate that a number of steps is desired. One of the issues is
> > that on machines like the KA10 and I7090 interrupts execute an
> > instruction out of sequence. Not sure if that should be counted as
> > a step or not.  
> 
> Well, if you want that behavior you can decrement sim_interval and
> \call sim_process_event and exit sim_inst() wherever you do that
> (being careful that everything is restartable from the point you
> exit).  Or reconsider the concept of having sim_interval track memory
> cycles instead of instructions.... What is the real value of keeping
> track of memory cycles?

  I track memory cycles so that I/O is closer to the time that the CPU
  expects. This is more important for the I7000 series since
  instructions where executed during I/O and some of the code relies on
  how many instructions can be executed during a read/write. Also for
  some cases during the I7090 I decrement sim_interval during long
  instructions to closer simulate the real speed of the machine. 

  When sim_inst() exits it is totally restartable. 

  This discussion might also apply to the Front Panel stuff and to
  throttling, to simulate closer to real machine speed. 

Rich

-- 
==========================================================================
Richard Cornwell
rich at sky-visions.com
http://sky-visions.com
LinkedIn:   https://www.linkedin.com/in/richard-cornwell-991076107
==========================================================================