Small IRA Tutorial

[kamelito]

The Vampire V4 support Emutos so I guess one could write an Atari app that hit the Amiga hardware is that also disabled?

[phx]

Yes. When IRA detects an Atari executable it will disable Amiga hardware symbols.

Theoretically I could make that an option, but who would ever seriously write Atari software, which only runs under EmuTOS on Amiga hardware?

[jotd]

IRA often detects wrongly Atari executables, ... then deletes the binary file probably because noone should use Atari.

[phx]

Fixed long ago. The -binary option always has priority now.

[hop]

Originally Posted by phx

During a long weekend (starting on Thursday) I found the motivation to implement loading of Atari TOS executables in IRA.

I'll grab your changes when I get some time and check you haven't broken Amiga

I may consider a quick look into Atari support for Aira if it would be of use to anyone in the community. I know nothing about the platform though, other than 68000 + AY chip + something called TOS!

Originally Posted by phx

Formatting is really broken in EAB since some time (maybe on non-Windows systems only?)

I can only see a single "P" in the first three CODE boxes unfortunately. Browsing in Windows 10 Chrome.

[phx]

Originally Posted by hop

I may consider a quick look into Atari support for Aira if it would be of use to anyone in the community.

It shouldn't be too much work. Call vasm with

-Ftos

instead of

-Fhunkexe

to generate the output. Autodetection of OS functions should be even easier, as you have a definite

trap

vector for BIOS, XBIOS and GEMDOS calls. The top word on the stack is the function code.

I can only see a single "P" in the first three CODE boxes unfortunately.

The single character was intended, because otherwise the code-block, which I used to separate paragraphs, would disappear.

Browsing in Windows 10 Chrome.

It's a Firefox problem with Preview Post.

[kamelito]

@Hop, Atari support would be great.

[kamelito]

@PHX it looks like that IRA does not support disassembly of *.library (I suppose then that its the same for devices) I tried directly into Airforce not the command line is that right? if yes why not?

Welcome to Aira Force V0.6.2

ira exited with code 0

ERROR: Invalid symbol name "LIBFUNC_$A"

ERROR: Failed to add symbol: SYMBOL LIBFUNC_$A $0003D1D0

ERROR: Failed to parse config directive on line 25: SYMBOL LIBFUNC_$A $0003D1D0

[hop]

I have found that ira can generate duplicate labels for jump tables with base address. This prevents the disassembled code from assembling. I've struggled to fix this locally and would be very grateful for any help.

Please find below a minimal repro.

Source code JumpTableBug.s:

; Minimal repro for jump table with base address label bug

	SECTION S_0,CODE

	moveq	#0,d0
	move.b	Status(PC),d0

	add.w	d0,d0 ; word offset
	move.w	CJumpTable(pc,d0.w),d0
	jmp	CJumpTable(pc,d0.w)

; C compilers often generate jump tables with entries relative to table address +2
CJumpTable: ; word sized elements
	dc.w	Red-(CJumpTable+2)
	dc.w	Green-(CJumpTable+2)
	dc.w	Blue-(CJumpTable+2)
;--------------------------------------------------------

Red:	
	move.w	#$f00,$dff180
	rts

Green:	
	move.w	#$0f0,$dff180
	rts

Blue:	
	move.w	#$00f,$dff180
	rts

Status:
	dc.b	2

	END

Assemble with vasm:

vasmm68k_mot -Fhunkexe -kick1hunks -no-opt -pic -wfail -warncomm -o JumpTableBug JumpTableBug.s

Process with ira config file JumpTableBug.cnf

MACHINE 68000
ENTRY $00000000
OFFSET $00000000
SYMBOL Status $00000034
SYMBOL CJumpTable $00000010
SYMBOL Red $00000016
SYMBOL Green $00000020
SYMBOL Blue $0000002A
CODE $00000000 - $00000010
JMPW $10 - $16 @ $12
CODE $00000016 - $00000034
END

with command line

ira -a -label=1 -radix=0 -compat=bi -config JumpTableBug

This generated JumpTableBug.asm has duplicate

CJumpTable

label, and table entries which don't reference the named symbol as expected:

; IRA V2.11 (Jun 19 2024) (c)1993-1995 Tim Ruehsen
; (c)2009-2024 Frank Wille, (c)2014-2019 Nicolas Bastien

COLOR00		EQU	$DFF180




	SECTION S_0,CODE

SECSTRT_0:
	MOVEQ	#0,D0			;00: 7000
	MOVE.B	Status(PC),D0		;02: 103a0030
	ADD.W	D0,D0			;06: d040
	MOVE.W	CJumpTable(PC,D0.W),D0	;08: 303b0006
	JMP	CJumpTable(PC,D0.W)	;0c: 4efb0002
CJumpTable:
CJumpTable:
	DC.W	(Red)-(LAB_10+2)	;10: 0004
	DC.W	(Green)-(LAB_10+2)	;12: 000e
	DC.W	(Blue)-(LAB_10+2)	;14: 0018
Red:
	MOVE.W	#$0f00,COLOR00		;16: 33fc0f0000dff180
	RTS				;1e: 4e75
Green:
	MOVE.W	#$00f0,COLOR00		;20: 33fc00f000dff180
	RTS				;28: 4e75
Blue:
	MOVE.W	#$000f,COLOR00		;2a: 33fc000f00dff180
	RTS				;32: 4e75
Status:
	DC.L	$02000000		;34
	END

I've managed to fix this, but I'm not happy with it. The fix results in two labels: the named symbol + the automatically generated label. I'm also not sure if there are any side effects. The patch is:

1. Modify

WriteLabel2

to pass

ira->p2labind

to

GetSymbol

instead of

index

2. Modify the final call to

GetSymbol

in

GetLabel

to pass

r

instead of

r2

This produces this, which isn't ideal, and I'm also concerned about side-effects:

...
	MOVE.W	CJumpTable(PC,D0.W),D0	;08: 303b0006
	JMP	CJumpTable(PC,D0.W)	;0c: 4efb0002
CJumpTable:
LAB_10:
	DC.W	(Red)-(CJumpTable+2)	;10: 0004
	DC.W	(Green)-(CJumpTable+2)	;12: 000e
	DC.W	(Blue)-(CJumpTable+2)	;14: 0018
Red:
...

Many thanks for any advice.

[phx]

Originally Posted by hop

The fix results in two labels: the named symbol + the automatically generated label.

That would be ok.

I'm also not sure if there are any side effects. The patch is:

1. Modify

WriteLabel2

to pass

ira->p2labind

to

GetSymbol

instead of

index

I would definitely expect side effects here. Because I did the change from

ira->p2labind

to

index

last December to fix something. It was done with this commit-log:

Bug fixes:
- Fixed TRAP_nn symbols for address range $80 to $bc.
- Instructions with PC-relative references outside of the current section
  are no longer ignored but generated with a constant displacement.
New features:
- Option -LABEL=<n> lets you select the label-style.
  0: original style with 4-digit continuous index: LAB_NNNN
  1: use hexadecimal address in label: LAB_ADDR
- Option -RADIX=<n> select the radix for displacements and equates.
  0: use hex for values >= 10, decimal for the rest (including negatives)
  10: print as decimal
  16: print as hexadecimal, except -9 to 9, which is printed decimal

Many thanks for any advice.

I have no good advice yet. These label issues are complex and I also need some debug time, because it is not my code.
But it must have something to do with symbols. Because with automatic labels the issue doesn't occur.

[hop]

Originally Posted by phx

That would be ok.


I would definitely expect side effects here. Because I did the change from

ira->p2labind

to

index

last December to fix something. It was done with this commit-log:

Bug fixes:
- Fixed TRAP_nn symbols for address range $80 to $bc.
- Instructions with PC-relative references outside of the current section
  are no longer ignored but generated with a constant displacement.
New features:
- Option -LABEL=<n> lets you select the label-style.
  0: original style with 4-digit continuous index: LAB_NNNN
  1: use hexadecimal address in label: LAB_ADDR
- Option -RADIX=<n> select the radix for displacements and equates.
  0: use hex for values >= 10, decimal for the rest (including negatives)
  10: print as decimal
  16: print as hexadecimal, except -9 to 9, which is printed decimal

I have no good advice yet. These label issues are complex and I also need some debug time, because it is not my code.
But it must have something to do with symbols. Because with automatic labels the issue doesn't occur.

Thanks very much. I am not familiar with this code either so it is slow going for me. It looks like that change in December may only result in the same label being written twice, but I may be wrong. It would be good to know why it was required. I'll debug too thanks.

[kamelito]

@PHX Archon do have many sections, Archon 2 even more. The code section are code is IRA able to disassemble code section as code by default ? I suppose that most C generated ASM contains code in code section. Doing it manually is cumbersome.