for ASM programmers

[meynaf]

On the Natami forums i have suggested some extensions for the N68050 (there i post under the name Phil G). Perhaps some will make their way in it, perhaps none.
I thought it could be good to somehow poll this forum's programmers for their needs.
So i ask you guys : what new features would you see to come on 68k ?
If you were designing it, what new instructions / addressing modes would you add, with which encoding ?

[Thorham]

Originally Posted by meynaf

If you were designing it, what new instructions / addressing modes would you add, with which encoding ?

None what so ever. What I want to see is the classic hardware recreated. Original specs with extras that aren't CPU/chip set related.

To me, 68K is fine the way it is

[Photon]

I'm with Thorham! even though if the chipset is something like a combo of CD32 and AGA with extra steroids, it would be interesting to see a fully compatible CPU remade running at the full speed possible today.

[Leffmann]

I'd like to see all addressing modes available as both source and destination, but I don't know how to solve the instruction encoding.

There's some good programming practice mumbojumbo behind f.ex not allowing PC-relative as destination, but I think if you're writing assembly you should know your stuff well enough not to screw up more than otherwise.

Another funny restriction in the M68K is that EOR from memory is not allowed, while AND and OR is. Why is that?

[Photon]

It could be that they ran out of encoding-bits left in the instruction word for that particular instruction or "instruction family" and thought it ugly to do a special separate instruction for that function. The encoding info in the programmer's manual could give a clue.

I should add that apart from a remade Amiga on steroids, I'm also all in favor of starting from the very base and get that perfect to build on that (Schoenfelt/A1k etc). Nothing wrong with that. But the ultimate dream would be to have a 100% compatible chipset and a completely remade but 100% compatible CPU at blinding speed (ie. super fast memory and running as fast as it can with excellent caches). That would open up some possibilities like fast web browsers and Javascript engines (HTML5 is the new black and Flash can go hide in shame, finally), and competent dev environments with nice high-level languages.

For other people of course, not me. I don't mind the limitations of the real deal, it's their challenge that is the reason for me to still be into Amigas.

Lordy, my English is starting to fail me, a clear signal that it's time for bed.

[meynaf]

When Mot' did 68000 they profiled real programs.

And then they've seen that EOR is less often used than the two other logic operations.

But this is IMO one of the big mistakes of 68k : once out of 2, when i need memory EOR it's not here !

There was enough encoding space to make full EOR (by using line-A space), but now it's too late - or it will be 2 bytes longer.

Other things that are missing are ability to make exg with memory, e.g. exg.b (a0),d0

The speed of the cpu is something else ; my question was above all about new stuff for programmers.

Now what ? Don't anyone have suggestions ?

[Wepl]

a mea (move effective address, e.g. mea (_xx,pc),(12,a0)) would be nice
but I doubt that any new instruction make sense, compatibility for user programs is enough

[Leffmann]

DBcc with longword size is probably something a lot of programmers have wished for.

[Samurai_Crow]

Originally Posted by Leffmann

DBcc with longword size is probably something a lot of programmers have wished for.

That's already implemented in the '050. Anything else?

[TCD]

Originally Posted by Samurai_Crow

That's already implemented in the '050. Anything else?

You mean it's already planned right?

[meynaf]

Originally Posted by Wepl

a mea (move effective address, e.g. mea (_xx,pc),(12,a0)) would be nice
but I doubt that any new instruction make sense, compatibility for user programs is enough

This one doesn't look like if it would be used quite often, but in fact you already have it if you want :

mea macro
 pea \1
 move.l (a7)+,\2
 endm

[pmc]

Originally Posted by Leffmann

DBcc with longword size is probably something a lot of programmers have wished for.

DBcc with longs would be nice. There is a trick that can be done to psuedo allow this though:

 
; d7 is the loop counter
 subq.l #1,d7
.loop: <loop code>
 
 DBcc d7,.loop
 Bcc .exit
 
 sub.l #$10000,d7
 bcc .loop
.exit:

Where the condition code in DBcc and Bcc above must be the same...

[Toni Wilen]

IMHO these are useless micro-optimizations that no one really needs, they only fragment Amiga hardware more and more..

[meynaf]

Fragmentation of Amiga HW is less important when you're gonna make a new machine anyway (with new features)... and this is the case with Natami.

[pandy71]

Some DSP instructions? inverse addressing (bit reverse)? hardware loop?

[Samurai_Crow]

Originally Posted by TheCyberDruid

You mean it's already planned right?

The 68050 is approaching final release and you still think it's in planning?

The 5th from the bottom of this page.

BTW, a bit reverse and byte reverse opcode are currently in there also.

[Samurai_Crow]

Originally Posted by pandy71

Some DSP instructions? -snip-

NatAmi's team is working on an actual DSP independently of the '050. Details of this processor are sketchy at this time though.

[TCD]

Originally Posted by Samurai_Crow

The 68050 is approaching final release and you still think it's in planning?

The 5th from the bottom of this page.

Was wondering why meynaf asked about stuff to add when the 68050 is already finished. So it's not, but that is already implemented. Okay.

[midnitoker]

Originally Posted by TheCyberDruid

Was wondering why meynaf asked about stuff to add when the 68050 is already finished. So it's not, but that is already implemented. Okay.

meynaf may be asking for future updates. Since it is all code, those can be downloaded and written to the Altera chip. And if I may nitpick - "approaching final" != final...

In any case, even if hardware and FPGA coding are near completion, the original 3.1 kickstart will still need some tweaks, for example adding USB support, bits to do with RTG, and surely the OS3.9 setpatch. I think all that won`t be ready too soon.

Why do I think so? As an ASM coder, and once part of the team, I was asked a couple of months back, but I`m a bit busy on AROS right now...

Anyone here up to that task? Sadly Gunnar seems to think that such tweaks can be hacked together in a week - he is a game-coder, which is linear, and not quite the same as complex, often interdependent and multitasking routines, that need to support (crash free!) the OS...

[TCD]

Originally Posted by midnitoker

And if I may nitpick - "approaching final" != final...

Yes and that's why I've written 'So it's not'. Info on the news page of the NatAmi site is a tad sparse and I don't follow the forum tbh. Hence I didn't know in which stage the 68050 was/is. Just wondered why this question (extensions for ther 68050) appears now that it 'approaches final' and not before. That's all.