[Coco] Is the 6309 worth it?

L. Curtis Boyle curtisboyle at sasktel.net
Wed Dec 6 16:11:10 EST 2006


On Wed, 06 Dec 2006 08:55:35 -0600, Roger Merchberger  
<zmerch-coco at 30below.com> wrote:
> There's 3 "modes" to the 6309 - 1) pure compatibility mode - to the
> software, it's a 6809. 2) "Enhanced 6809" mode: no new instructions, but
> some instructions take fewer cycles to complete - that's the 15% number
> you've heard. That can goof up some timing loops in games & whatnot. 3)
> full 6309 Mode: There are extra registers, extra instructions, extra,
> extra, extra and some instructions execute even quicker, IIRC.[1]

     Actually, you can use the extra instructions without being in "native  
mode", as most of us call it. The Burke&Burke 6309 patches for OS9  
(original version, anyways), ran in compatibility mode, but with the extra  
instructions. The main differences between native and compatibility mode  
is that the pre-fetch byte is cached, so a lot of instructions are faster  
by a CPU cycle, and that you're IRQ's now push the new W register onto the  
stack as well as the 8 regular registers.

>
> You get 2 extra 8-bit accumulators, which can also be utilized as a  
> 16-bit
> accumulator, and *also* with the existing 16-bit accumulator (D) to make  
> a
> 32-bit accumulator{!} amongst other things.
>
> I dunno about the "10x faster" bit, but software written to take  
> advantage
> of the 6309 can be quite snappy, comparatively (yet vaguely ;-) speaking.

     Yes, some things, particularily block memory moves, can get up to 10x  
faster. This can be used for things as diverse as copying a disk buffer to  
scrolling or clearing a screen. Also, since no hardware divide existed at  
all on the 6809, using the new hardware divide (16/8 or 32/16) is around  
10x faster than doing it through software routines.

>>  It seems like a 32 mhz clock crystal (2.0/4.0
>> mhz) would do the same thing, and be less sensitive, but what about
>> compatibility?
>
> It wouldn't do "the same thing." The 6309 gains speed due to efficiency,  
> a
> crystal swap gains speed via overclocking, and remember: you're
> overclocking the system buss. So, not only will your CPU overheat  
> (because
> dollars-to-donuts it can't dissipate that much heat with convective
> cooling) you're also overclocking the memory (which won't be  
> none-too-happy
> about that!) and the GIME (which quite likely will gesplode) and of  
> course,
> the video circuitry, which means you won't be able to see how fast your
> computer's running before it fries.
    I can honestly say that the 32MHz crystal, except for screwing up  
printer and cassette timing loops, has never caused my 6809 or 6309's to  
overheat much (and never to cause excessive crashing). I have ran 4  
different machines with that upgrade, with both 6809's and 6309's. The  
best case is to put in the crystal upgrade, use a 6309, and run Nitros9  
(but I would be a bit biased that way, wouldn't I? <Grin>
     The biggest hurdle of using a 6309 is if you use RS-DOS almost  
exclusively... there are only a few programs that properly use it (Sock  
Master's MOD player, Roger's Spyder Hyper & Projector III, and maybe a few  
others. I believe Robert Gault came out with some patches for the Coco 3  
only to allow RS-DOS to run in native mode, getting you that 12-15% speed  
increase, and I know Sock Master had some patches for screen scrolling as  
well, that helped a lot, but not much more than that.

-- 
L. Curtis Boyle



More information about the Coco mailing list