Preservation of registers

[guy lateur]

I think I read somewhere that d0 & d1 and a0 & a1 can be changed by subroutines, while the rest is supposed to be left unchanged. Is this correct? For example:

;-- copy chars into textBuffer and zero-terminate it
;in:    a0:     source pointer
;in:    d0:     # characters to copy
copyToBuffer:
    move.l  4.w,a6
    lea     textBuffer,a1
    jsr     _LVOCopyMem(a6)     ;does CopyMem() preserves a1?
    lea     textBuffer,a1       ;if it doesn't: reset it
    add.l   d0,a1
    move.b  #asciiNULL,(a1)
.return:
    rts

If a1 is changed by the CopyMem() call, I'll need to reset it before I can write the zero-terminator. Well, actually, in this case, it would probably already point to the place where the NULL should be written, so I could probably leave out the reset & the add.

But you know what I mean. I'm asking because this doesn't seem to be described in the docs (http://amigadev.elowar.com/read/ADCD.../node01F9.html). So what are the default rules and assumptions for this?

[roondar]

It’s hidden away a bit, but you can read about the library function register use here: http://amigadev.elowar.com/read/ADCD.../node000F.html

The short version is that Amiga library functions use d0/d1 and a0/a1 and expect the library pointer to be in a6. All of these apart from a6 can end up trashed. Other registers are safe.

Edit: made the above say what I meant

[phx]

The usual ABI for most 68k operating systems implies that d0, d1, a0, a1 are scratch registers, which are not restored after function calls (Atari TOS also trashes d2 and a2, IIRC). This is true for all functions in AmigaOS, and for all functions generated by Amiga C compilers.

When calling your own functions in assembler you can make your own rules, but generally it is a good idea to always follow the ABI, so you can depend on it everywhere in your code and don't have to check every function you call.

[alkis]

OS-calls treat, as you wrote, d0/d1/a0/a1 as scratch registers. You can't expect their values to be preserved, UNLESS it is specifically stated in the autodocs. (for example look at Results in this one http://amigadev.elowar.com/read/ADCD.../node0339.html)

[phx]

Originally Posted by roondar

All apart from a6 can end up trashed.

Er... no!
d2-d7 and a2-a7 are guaranteed to be preserved by all OS-calls.

[roondar]

Originally Posted by phx

Er... no!
d2-d7 and a2-a7 are guaranteed to be preserved by all OS-calls.

That is what I said, the ‘all’ I referred to was supposed to indicate the four registers I pointed out

Edit: reading it back I do understand why you thought I meant differently. My mind was ahead of me again...

[phx]

Originally Posted by roondar

That is what I said, the ‘all’ I referred to was supposed to indicate the four registers I pointed out

Ah...ok - now as I reread your sentence. But it could easily be misunderstood.

[roondar]

Originally Posted by phx

Ah...ok - now as I reread your sentence. But it could easily be misunderstood.

You are correct, I have edited the post so it’s clear now.

[guy lateur]

Originally Posted by roondar

You are correct, I have edited the post so it’s clear now.

Thanks for clearing this up, guys!

And while on the subject of clarity: the libray docs state "Amiga library functions use registers D0, D1, A0 and A1 for work space and use register A6 to hold the library base. Do not expect these registers to be the same after calling a function."

At first, I honestly also read that as "Do not expect any of the aforementioned registers..", and therefor "A6 may also be thrashed". But we've established it's only d0/d1 & a0/a1, right?

Thanks for all the feedback, people, much appreciated!

[roondar]

AFAIK a6 doesn’t change.

[guy lateur]

Originally Posted by roondar

AFAIK a6 doesn’t change.

That would definitely make the most sense, alright.

I'm not really sure why it is even mentioned in that part of the docs pertaining to preserved registers, to be honest. Unless the library expects you to set its own base adress there? That's not the case though, is it? I mean, I could just as well call the functions using a5 to contain the base, right?

[Don_Adan]

Originally Posted by guy lateur

That would definitely make the most sense, alright.

I'm not really sure why it is even mentioned in that part of the docs pertaining to preserved registers, to be honest. Unless the library expects you to set its own base adress there? That's not the case though, is it? I mean, I could just as well call the functions using a5 to contain the base, right?

For most library calls A6 as base is not necessary, but for some calls is necessary. And your code will be crashed if A5 will be set as library base. Clean code must have always A6 as library base.

[Galahad/FLT]

Originally Posted by roondar

AFAIK a6 doesn’t change.

I don't think thats entirely true.

If you open the dos.library and want to then access functions of that library, it expects you to put the dos.library base address into A6.

Obviously it won't do this by itself, its something you have to specifically do, but just to point out there there are some scenarios where A6 can and has to change and you have to remember that if you then call a different library or go back to exec.

[guy lateur]

Originally Posted by Don_Adan

For most library calls A6 as base is not necessary, but for some calls is necessary. And your code will be crashed if A5 will be set as library base. Clean code must have always A6 as library base.

I see, now that's interesting. Although I've always done this, I didn't know it was a requirement in some cases, so thanks for pointing that out! Can you maybe think of an example where this is the case?

[guy lateur]

Originally Posted by Galahad/FLT

I don't think thats entirely true.

If you open the dos.library and want to then access functions of that library, it expects you to put the dos.library base address into A6.

Obviously it won't do this by itself, its something you have to specifically do, but just to point out there there are some scenarios where A6 can and has to change and you have to remember that if you then call a different library or go back to exec.

Really? So you're saying that sometimes:
1. the base needs to be in a6
2. a6 may have changed after the call
That would explain the way it's frased in the docs, though.

[Don_Adan]

Originally Posted by guy lateur

I see, now that's interesting. Although I've always done this, I didn't know it was a requirement in some cases, so thanks for pointing that out! Can you maybe think of an example where this is the case?

A6 is library base, at plus offset is base data. Some calls reads data from library base, if you call library with other register as base, library routine will be reads data not from library data, but from random A6 value. Anyway something like this is safe to use:

move.l A6,A5
jsr -$xx(a5)

but no big sense for me.

[Don_Adan]

Originally Posted by Galahad/FLT

I don't think thats entirely true.

If you open the dos.library and want to then access functions of that library, it expects you to put the dos.library base address into A6.

Obviously it won't do this by itself, its something you have to specifically do, but just to point out there there are some scenarios where A6 can and has to change and you have to remember that if you then call a different library or go back to exec.

From my attempt to reworking of some rom libraries/devices etc. A6 is always unchanged after all known me library/device calls. Of course often is changed in middle code, but is always returned to original A6 input value.

[guy lateur]

Originally Posted by Don_Adan

A6 is library base, at plus offset is base data. Some calls reads data from library base, if you call library with other register as base, library routine will be reads data not from library data, but from random A6 value. Anyway something like this is safe to use:

move.l A6,A5
jsr -$xx(a5)

but no big sense for me.

Maybe my question wasn't clear. I understand what the problem is, I was just wondering if you happened to know of any specific library call (eg, like OpenDevice() in exec lib) that expects the base to be in a6. Anyway, it's not that important; I will definitely have my bases in a6, from now on. All bases are belong to a6!

[PeterK]

add.l d0,a1

the contents of d0 is unknown here, too. This could cause random memory trashing instead of terminating your text string.

[Don_Adan]

Originally Posted by guy lateur

Maybe my question wasn't clear. I understand what the problem is, I was just wondering if you happened to know of any specific library call (eg, like OpenDevice() in exec lib) that expects the base to be in a6. Anyway, it's not that important; I will definitely have my bases in a6, from now on. All bases are belong to a6!

Yes, I know. You can check DOS.lzx from:

http://wt.exotica.org.uk/test.html