Compile and link vs. whatever zcc does

[derekfountain]

I have a piece of test code which I've distilled to this:

Code: Select all

#pragma output REGISTER_SP = 0xD000

#include <arch/zx.h>
#include <arch/zx/sp1.h>

int main()
{
  sp1_Initialize( SP1_IFLAG_MAKE_ROTTBL | SP1_IFLAG_OVERWRITE_TILES | SP1_IFLAG_OVERWRITE_DFILE,
                  INK_BLACK | PAPER_WHITE,
                ' ' );
  
  while(1) {
    zx_border(INK_BLACK);
    zx_border(INK_WHITE);
  }
}

I can build a TAP with this:

Code: Select all

zcc +zx -vn -startup=31 -clib=sdcc_iy sp1_test.c -o sp1_test -create-app

which works fine. The sp1 call does practically nothing and the border goes fuzzy.

If I build it with these two lines:

Code: Select all

zcc +zx -compiler sdcc -c sp1_test.c -o sp1_test.o -I$ZCCCFG/../../include/_DEVELOPMENT/sdcc
zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -create-app

then I'm not sure what happens. The machine locks up (expected since there's an infinite loop) but the border doesn't go fuzzy so I'm presuming the sp1 call is actually doing locking.

I've tried various simple permutations of files and library calls, but they all seem OK. It's the use of the SP1 library which seems sensitive to whatever I'm doing wrong. Looking carefully at the compile lines (with -v output) I can see subtle differences between the two approaches. The nuances are rather beyond me though.

What's the correct compiler command line for building individual C sources for linking?

[dom]

In your split build you're creating the .o file with classic and then linking the final library with newlib.

So, you need to add -clib=sdcc_iy to the line that creates the .o

[alvin]

If I build it with these two lines:

Code: Select all

zcc +zx -compiler=sdcc -c sp1_test.c -o sp1_test.o -I$ZCCCFG/../../include/_DEVELOPMENT/sdcc
zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -create-app

then I'm not sure what happens. The machine locks up (expected since there's an infinite loop) but the border doesn't go fuzzy so I'm presuming the sp1 call is actually doing locking.

This one is mixing up the classic library and the newlib.

zcc +zx -compiler=sdcc -c sp1_test.c -o sp1_test.o -I$ZCCCFG/../../include/_DEVELOPMENT/sdcc

is doing a classic compile (no clib=...). I think your include path is overriding the classic library's so that it is finding the right prototypes and formatting the calls properly. But this classic compile is an sdcc_ix compile so I'm not certain exactly what happens.

What's the correct compiler command line for building individual C sources for linking?

Start with the correct compile line:

Code: Select all

zcc +zx -vn -startup=31 -clib=sdcc_iy sp1_test.c -o sp1_test -create-app

And then tell zcc to stop at object file by adding "-c":

Code: Select all

zcc +zx -vn -c -startup=31 -clib=sdcc_iy sp1_test.c -o sp1_test -create-app

This will work but the output object file will be "sp1_test" because zcc is being told to rename it by "-o". The create-app step is never taken because a binary is not being produced so that option can be removed. And since no binary is being made, the startup to select a crt is unncessary too. The compilation is stopped before the binary building step.

So the correct invoke is:

Code: Select all

zcc +zx -vn -c -clib=sdcc_iy sp1_test.c

[alvin]

It looks like I doddle a lot more tan dom

[derekfountain]

So that would give me these lines:

Code: Select all

zcc +zx -compiler sdcc -clib=sdcc_iy -c sp1_test.c -o sp1_test.o -I$ZCCCFG/../../include/_DEVELOPMENT/sdcc
zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -create-app

There's no (apparent) difference in the behaviour of the final program.

[alvin]

try this:

Code: Select all

zcc +zx -v -c -clib=sdcc_iy sp1_test.c
zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -create-app

Make sure the old objects are gone just in case an error is occurring and it's picking up an old .o

[derekfountain]

OK, catching up with overlapping posts, I tried Alvin's lines:

Code: Select all

zcc +zx -vn -c -clib=sdcc_iy sp1_test.c
zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -create-app

Still broken. But if I comment out the sp1 init call those build a working app.

[alvin]

You're right. Something odd there - I'm looking into it now. The object linked one is sitting on a halt with ints disabled.

[alvin]

The pragma is not active in the object file compile version so that the stack is not initialized at address 0xd000.

Code: Select all

zcc +zx -vn -c -clib=sdcc_iy sp1_test.c

This sees the pragma and creates the file "zcc_opt.def" to store the initial stack location. Pragmas are communicated to the crt when the binary is built. BUT:

Code: Select all

zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -create-app

Starts a new compile and erases the "zcc_opt.def" file. So the stack pragma is lost.

There is a way to tell zcc to preserve the zcc_opt.def file so that pragmas seen in separate compilation are accumulated (see the -preserve option for zcc) but it is much cleaner and simpler to store your pragmas in a separate pragma file:

zpragma.inc

Code: Select all

#pragma output REGISTER_SP = 0xD000

Then the compile line forming the binary reads the pragma file:

Code: Select all

zcc +zx -vn -c -clib=sdcc_iy sp1_test.c
zcc +zx -v -startup=31 -clib=sdcc_iy sp1_test.o -o sp1_test -pragma-include:zpragma.inc -create-app

[alvin]

I guess the best way to see what pragmas were active in a compile is to take a look at the zcc_opt.def file after the binary is built. In this case you would have noticed the REGISTER_SP pragma was missing from the file.

[derekfountain]

OK, my makefile is now working with the external pragma file. Thanks!