C2P....help!

[NovaCoder]

I'm currently trying to stick someone's assembler C2P routine into my C++ project, I think I've now got the bugger to link properly but it's not actually working yet.


As I'm crap at assembler can someone have a look at the attached routines and tell me (a) what I should be passing as arguments and (b) if any better routines exist.

http://home.iprimus.com.au/novacoder/c2p_030.h
http://home.iprimus.com.au/novacoder/c2p_030.s
http://home.iprimus.com.au/novacoder/c2p_030.o

The init method seems to work ok (as in it doesn't crash).

extern "C" void c2p1x1_cpu3blit1_queue_init (
                      UWORD chunkyx,  // width
                      UWORD chunkyy,  // height
                      UWORD scroffsy, // y offset
                      ULONG bplsize,  // modulo to next plane
                      ULONG signals1,
                      ULONG signals3,
                      struct Task *thistask,
                      struct Task *other

But this sucker just crashes my system though (causes a restart!).

extern "C" void c2p1x1_cpu3blit1_queue (
                      UBYTE *c2pscreen,
                      UBYTE *bitplanes);

This was borrowed from ADoom BTW.

Thanks

[NovaCoder]

In particular I'm having trouble with the c2pscreen argument, not sure what that should be.

[korruptor]

Pointer to the chunky buffer?

[NovaCoder]

Originally Posted by korruptor

Pointer to the chunky buffer?

Well that's what I assumed but it just hangs.

I'm doing this for a 320x200 bit bit chunky display:

 static byte *backBuffer;
 
 backBuffer=(byte*)malloc(64000);

The back buffer is then used by the following method:

void I_FinishUpdate ()
{
  int top, left, width, height;
  total_frames++;
 

    left = 0;
    top = 0;
    width = SCREENWIDTH;
    height = SCREENHEIGHT;

  if (video_is_native_mode) {
    if (video_is_using_blitter) {
      start_timer();
      Wait (1 << video_sigbit1); /* wait for prev c2p() to finish pass 3 */
      blit_time += end_timer();
      if (video_blit_is_in_progress) {
   start_timer();
   Wait (SIGBREAKF_CTRL_F); /* wait for prev ChangeScreenBuffer() safe */
   safe_time += end_timer();
   video_blit_is_in_progress = FALSE;
      }
      video_which = 1 - video_which;  /* render to the hidden bitmap */
      start_timer();
      c2p1x1_cpu3blit1_queue (backBuffer, video_raster[video_which]);
      c2p_time += end_timer();
      video_blit_is_in_progress = TRUE;
    }
  }
}

And it crashes on the call to 'c2p1x1_cpu3blit1_queue'.

[Kalms]

The C code sends the arguments on the stack, but the assembly code expects them in registers.

Either

A) change the C function prototype to specify register locations for all arguments (exact syntax is compiler dependent)

or

B) change the assembly code to load register contents from the stack.

[NovaCoder]

Originally Posted by Kalms

The C code sends the arguments on the stack, but the assembly code expects them in registers.

Either

A) change the C function prototype to specify register locations for all arguments (exact syntax is compiler dependent)

or

B) change the assembly code to load register contents from the stack.

Yep that seems to be the problem, the code below *might* work for straight c but won't compile in c++

#ifndef __GNUC__
#define REG(reg,arg) register __##reg arg
#else
#define REG(reg,arg) arg __asm(#reg)
#endif
 
 
extern void c2p1x1_cpu3blit1_queue_init (
REG(d0, UWORD chunkyx), // width
REG(d1, UWORD chunkyy), // height
REG(d3, UWORD scroffsy), // y offset
REG(d5, ULONG bplsize), // modulo to next plane
REG(d6, ULONG signals1),
REG(d7, ULONG signals3),
REG(a2, struct Task *thistask),
REG(a3, struct Task *othertask),
REG(a4, UBYTE *chipbuff));
 
extern void c2p1x1_cpu3blit1_queue (
REG(a0, UBYTE *c2pscreen),
REG(a1, UBYTE *bitplanes));

[NovaCoder]

Salass00 also come up with this idea:

If you can't use register arguments (IIRC C++ doesn't support this) then you can create small assembler stub functions that take arguments on the stack and put them in the correct registers like this:

XREF _c2p1x1_cpu3blit1_queue_init 
XREF _c2p1x1_cpu3blit1_queue 
XDEF _c2p1x1_cpu3blit1_queue_init_stub 
XDEF _c2p1x1_cpu3blit1_queue_stub 

_c2p1x1_cpu3blit1_queue_init_stub: 
movem.l 4(a7),d0-d1/d3/d5-d7/a2-a4 
jsr _c2p1x1_cpu3blit1_queue_init 
rts 

_c2p1x1_cpu3blit1_queue_stub 
movem.l 4(a7),a0-a1 
jsr _c2p1x1_cpu3blit1_queue 
rts

Then in your C++ code you use these two functions instead:

extern "C" void c2p1x1_cpu3blit1_queue_init_stub (UWORD chunkyx, UWORD chunkyy, UWORD scroffsy, ULONG bplsize, ULONG signals1, ULONG signals3, struct Task *thistask, struct Task *othertask, UBYTE *chipbuff);

extern "C" void c2p1x1_cpu3blit1_queue_stub (UBYTE *c2pscreen, UBYTE *bitplanes);

[Kalms]

Originally Posted by NovaCoder

_c2p1x1_cpu3blit1_queue_init_stub: 
movem.l 4(a7),d0-d1/d3/d5-d7/a2-a4 
jsr _c2p1x1_cpu3blit1_queue_init 
rts

I presume the calling conventions of your C/C++ compiler is that d0-d1/a0-a1 are scratch, and others should be preserved by the called function. If so, the above stub will clobber some registers for you. This will cause mysterious crashes in optimized builds in the nearby region after the init call.

The stub below preserves registers properly:

_c2p1x1_cpu3blit1_queue_init_stub: 
movem.l d2-d7/a2-a6,-(sp)
movem.l 11*4+4(a7),d0-d1/d3/d5-d7/a2-a4 
jsr _c2p1x1_cpu3blit1_queue_init 
movem.l (sp)+,d2-d7/a2-a6
rts

[NovaCoder]

For other's reference, this got it working (thanx Salass00)

	XREF _c2p1x1_cpu3blit1_queue_init
	XREF _c2p1x1_cpu3blit1_queue
	XDEF _c2p1x1_cpu3blit1_queue_init_stub
	XDEF _c2p1x1_cpu3blit1_queue_stub

_c2p1x1_cpu3blit1_queue_init_stub:
	movem.l d3/d5-d7/a2-a4,-(a7)
	movem.l 32(a7),d0-d1/d3/d5-d7/a2-a4
	jsr _c2p1x1_cpu3blit1_queue_init
	movem.l (a7)+,d3/d5-d7/a2-a4
	rts

_c2p1x1_cpu3blit1_queue_stub:
	movem.l 4(a7),a0-a1
	jsr _c2p1x1_cpu3blit1_queue
	rts