z88dk forums

Hi Philipp,

Which features necessary or useful for use of sdcc in z88dk do you
consider the most important?

For ease of drop-in the following are essential:

- SDCC generates an invalid pseudo-op at the top of the assembler file (.opt)
- Reverse parameter parsing (i.e. left to right rather than right-to-left) with a containing the number of words on the stack for variadic functions - possibly as a function decorator to allow a gradual migration to right-to-left convention

For compact code:

- Change initialisation of static data so that it's contained within defb rather than initialised with ld (xx),hl

Without these fixes we can't connect the z88dk libraries with the sdcc compiler.

* --fomit-frame-pointer (necessary when ix cannot be used) is still
somewhat broken.
* --reserve-regs-iy (necessary when iy cannot be used) seems to work well.

We can initially cope with only one index register being used - for certain platforms we can swap them round in the assembler, I think there's only a couple of platforms that prohibit use of either of them.

With these 3 changes in place, we should then be able to offer sdcc as a drop in compiler whilst maintaining compatibility with the majority of our existing libraries and most projects apart from those that use floating point.



------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

Am 07.06.2011 19:38, schrieb Stefano Bodrato (stefano_bodrato@...):

Dom, thank you for spotting them out, I'm adding a comment, hope it
doesn't add complexity 

For compact code: - Change initialisation of static data so that
it's contained within defb rather than initialised with ld (xx),hl

z88dk just links a 'main' program, then jumps to it. If I remember
well the sdcc main entry name not "main". Perhaps a possible
workaround could be to label the init portion with 'main' and make it
end with a direct jump in "sdcc main".

wouldn't that be linking in another object file containing:

main:
jp _main

and thus something that's easy to do outside the compiler?

Philipp

------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

Am 07.06.2011 11:34, schrieb dom (dom@...):

- Reverse parameter parsing (i.e. left to right rather than
right-to-left) with a containing the number of words on the stack for
variadic functions - possibly as a function decorator to allow a
gradual migration to right-to-left convention

Most of it doesn't work yet, but it may be enough for what you want to do.

Things that don't work: Function definitions, function pointers.
Thus you can now call functions from sdcc passing parameters in reverse
order as long as you don't use function pointers, but you can't have an
sdcc function accept it's parameters in reverse order yet.

void f(int, int) __optlink;
void h(int, int);
void (*p)(int, int) __optlink;
void (*q)(int, int);

void g(void)
{
f(1, 2); // OK - parameters passed in reverse order
h(1, 2); // OK - parameters passed in normal order
(*p)(1, 2); // Broken
(*q)(1, 2); // OK
}

Philipp


------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 07.06.2011 11:34, schrieb dom (dom@...):

- Change initialisation of static data so that it's contained within
defb rather than initialised with ld (xx),hl

How important is this? One one hand it's purely a code size issue, on
the other hand code size matters (and was one of the main motivations
for the new register allocator in sdcc).
For me, such data typically is only a few bytes per program. Does anyone
really use large amounts of static data?

Philipp
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3vb7AACgkQbtUV+xsoLpp3/gCfbH+I9GkqpL1aI9KmOxSg+CPl
UrMAoN18+k0xqUPV+VhkCBquYgC9N1wd
=E6Hl
-----END PGP SIGNATURE-----

------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 08.06.2011 15:15, schrieb dom (dom@...):

Am 07.06.2011 11:34, schrieb dom (dom@...):

- Change initialisation of static data so that it's contained
within defb rather than initialised with ld (xx),hl

How important is this? One one hand it's purely a code size issue,
on the other hand code size matters (and was one of the main
motivations for the new register allocator in sdcc). For me, such
data typically is only a few bytes per program. Does anyone really
use large amounts of static data?

While I think that sdcc should support using defb for static one day it
will probably take some time; until then, I suggest to fix applications
that suffer from the current situation.

I think it depends on what you're writing. One of our examples
adv_a.c contains a large amount of static data and I can envisage
games in general using a large amount of static data, eg sprites,
room layouts etc. One method that I've seen quite a lot is for the
graphics (or whatever) tool to export the data as char array. As an
example, one of my old games has 80 sprites of 36 bytes each. Which
is, give or take, 3k, if not statically initialised (or merged in
another way) that goes up to around 10k!

As long as these are never modified there should be no problem. Global
variables that are not modified (as all that stuff in adv_a.c, or
sprites, graphics, etc in my games) can be placed in the _CODE segment
by sdcc, where defb is used. If you tell sdcc that it will never be
modified:

unsigned char x = 7;
const unsigned char y = 7;

becomes

.area _GSINIT
;test.c:1: unsigned char x = 7;
ld    iy,#_x
ld    0 (iy),#0x07
.area _CODE
_y:
.db #0x07    ; 7

The other reason is the inability of z80asm to stitch together the
gsinit sections and so multi-file projects fail to initialise
properly.

That looks like a bigger problem.

Philipp

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3vfBUACgkQbtUV+xsoLpqK0gCgmwfI1gVPkMvyh5RgRUltAj7X
G2AAoIi8JJAjVnea/pIB53pH7LYcpDve
=I0Es
-----END PGP SIGNATURE-----

------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

Am 07.06.2011 11:34, schrieb dom (dom@...):

- Reverse parameter parsing (i.e. left to right rather than
right-to-left) with a containing the number of words on the stack for
variadic functions - possibly as a function decorator to allow a
gradual migration to right-to-left convention

As far as I can see, z88dk expects the number of parameters in register
A when using varargs. Is this correct?

Currently __optlink will reverse the order of parameters for varargs,
but doesn't communicate the number of parameters to the callee.

I think that will be okay, the only varargs functions we've got are the *scanf and *printf and some syscall functions for the Rex6000. Since the stdio functions are written in C, I think that's pushed me towards having a different set of libraries for sdcc builds with the C functions being right-to-left and the assembler ones left-to-right.



------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 07.06.2011 11:34, schrieb dom (dom@...):

- Reverse parameter parsing (i.e. left to right rather than
right-to-left) with a containing the number of words on the stack for
variadic functions - possibly as a function decorator to allow a
gradual migration to right-to-left convention

Do you happen to have a name for this calling convention? It has been
brought to my attention, that "__optlink" is a bad name for this.

Philipp
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3v2lkACgkQbtUV+xsoLpo9WACfW5pTrxn5Ye9sEqcAJ1+Cgc+r
hS4AoLj/2fdhTE5MLDEEJQturFK24Fhy
=+joG
-----END PGP SIGNATURE-----

------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 08.06.2011 22:30, schrieb dom (dom@...):

Do you happen to have a name for this calling convention? It has
been brought to my attention, that "__optlink" is a bad name for
this.

Looks like it's called "pascal" convention in the Visual Studio
world:
http://blogs.msdn.com/b/oldnewthing/arc … 47184.aspx

The pascal calling convention is left-to-right, callee-cleans.
The convention I implemented is left-to-right, caller-cleans.

Philipp
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3v3cQACgkQbtUV+xsoLpr6iACg4kxePUCHY4qo6OfURwvc8onW
920AoItYUIzgcWNcRZ25p69IAMZtfvG6
=6ic8
-----END PGP SIGNATURE-----

------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev

Am 08.06.2011 22:46, schrieb dom (dom@...):

The pascal calling convention is left-to-right, callee-cleans. The
convention I implemented is left-to-right, caller-cleans.

Close! Why don't just call it __smallc since that's what all the
SmallC compilers generate? Or is that too specific?

Well, __smallc is probably the best choice. I was just wondering if
maybe this convention is more widespread, so it has a generic, more
well-known name.

Philipp

------------------------------------------------------------------------------
EditLive Enterprise is the world's most technically advanced content
authoring tool. Experience the power of Track Changes, Inline Image
Editing and ensure content is compliant with Accessibility Checking.
http://p.sf.net/sfu/ephox-dev2dev