[Sdcc-user] missing push / pop pair

[Kai V. <volkmar@s.netic.de>]

Hi Sandeep, hi all,

at first I'd like to thank you Sandeep and everybody involved in the
sdcc project for
doing such great work. I have been searching many times for an 8051
compiler
which is comparably fast as commercial products but being open-source.

Developing a medium size project I was forced to program an
Intel-Hex-File
reading boot loader (utilizing one of the serial interfaces) which may
be used
to download the functional code into flash program memory.

As a first step I tried to design an Intel-Hex-reading function. Because
this
file format is very simple I didn't expect much difficulties...
I was surprised that the checksum accumulation didn't work at all. It
seemed
as if there were "swallowed" bytes. To debug this code I used already
made
serial I/O-routines which enabled me to use one RS232 to perform debug
output
while the other is bound to the Intel-Hex supplying terminal.
I discovered the astonishing fact that the code worked properly if the
debugging output was made (line 76) and didn't work if the "production
version" of the code was compiled (without the debugging output).
By comparing the list files of o.k. and defective code I found that the
call
of the serial output function not only changed the nearby code but the
following, too. This may be the reason why the checksum calculation is
corrupted if no serial output is made.
I enclosed the important part of the function:


		/* get higher byte of first address */
		databyte = monitor_readbyte();
		checksum += databyte;
		serial0_put_hex_byte(checksum);								/*!!! may not be commented
out!!!*/
 		address = (unsigned int)databyte << 8;

		/* get lower byte of first address */
		databyte = monitor_readbyte();
		checksum += databyte;
		address += databyte;



If the marked line (76) /* !!! ... !!!*/ is included, the following code
fragment is generated:

                            523 ;	monitor.c 75
   0074 EB                  524 	mov	a,r3
   0075 2A                  525 	add	a,r2
   0076 FC                  526 	mov	r4,a
                            527 ;	monitor.c 76
   0077 C0 02               528 	push	ar2
   0079 C0 03               529 	push	ar3
   007B C0 04               530 	push	ar4
   007D 8C 82               531 	mov	dpl,r4
   007F 12s00r00            532 	lcall	_serial0_put_hex_byte
   0082 D0 04               533 	pop	ar4
   0084 D0 03               534 	pop	ar3
   0086 D0 02               535 	pop	ar2
                            536 ;	monitor.c 80
   0088 C0 02               537 	push	ar2
   008A C0 04               538 	push	ar4
   008C 12s00r07            539 	lcall	_monitor_readbyte
   008F AB 82               540 	mov	r3,dpl
   0091 D0 04               541 	pop	ar4
   0093 D0 02               542 	pop	ar2

If the line (76) is commented out, it looks like:

                            523 ;	monitor.c 75
   0074 EB                  524 	mov	a,r3
   0075 2A                  525 	add	a,r2
   0076 FC                  526 	mov	r4,a
                            527 ;	monitor.c 80
   0077 C0 02               528 	push	ar2
   0079 12s00r07            529 	lcall	_monitor_readbyte
   007C AB 82               530 	mov	r3,dpl
   007E D0 02               531 	pop	ar2

You may see, the code originating from line 80 lacks a push/pop pair
which
saves and restores ar4 onto/from the stack. The function
monitor_readbyte
(which is contained in the same module) uses and destroys ar4.

I'm using sdcc 2.2.0 (Linux) which I got from the CVS tree today (July
25th).
The compiler was invoked by
	sdcc -c monitor.c -I /root/micro/sdcc/device/include/ --code-loc 0x8000
	--xram-l oc 0x0000 --data-loc 0x30 --stack-after-data
	-L//D/usr/local/share/lib/small -I//D/usr/local/share/include/
without any additional #pragmas.

I'll try to work_around the problem but would be glad if somebody could
find the
problem.

Greetings from Germany

Kai