The assembler


XA supports both the standard 6502 opcodes as well as the CMOS versions (Rockwell 65c02). Not supported are the 6502 undocumented opcodes, they have to be put in by hand (with ".byte" directives).

NOTE: This version of XA is based from the original source code when the development was abandonned around 1998.
Since then, the development of the official version had started again and then stopped again.

You can find the other official releases (2.3.x) on these two places:

Parameters and features

The assembler contains only a single programm called "xa". It takes one or more Source files into one object file, that can directly be used. But the assembler also has a mode to produce relocatable files, conforming to the 'o65' fileformat (See fileformat.txt).


xa [options] Source1 [Source2 ...] 
Object: this is the name, the output (object) file gets Error: Here you will find the Error listing. Label: this is the label list

'-C' 		gives error codes when using CMOS-opcodes. Default is not to complain.
'-c'		do not produce o65 executable, but object files that can contain undefined references.
'-cc'       almost identical to '-c' except the result is usable with CC65 tools.
'-v' 		go into verbose mode
'-R' 		do not produce absolute code, but do relocation and all that.
'-o filename'	set output filename
'-e filename'	set errorlog filename
'-l filename'	set labellist filename
'-r'		add crossreference list to labellist output (i.e list of filename/line where label is used)
'-M'		This option used to allow ':' to appear in comments after a semicolon (MASM mode), but this 
            is now the default behavior.
'-b? adr'	set segment start address for ? = t(ext), d(ata), b(ss) or z(ero) segment.
'-A adr'    the _file_ starts at adr in a ROM, then the text segment need not be relocated. 
            That of course only works, if the data/bss/zero segments are not occupied by other programs.
'-G'		omit writing the exported globals to the file.
'-B'		Show lines with '.(' or '.)' pseudo opcodes
'-Llabel'	defines 'label' as absolute, undefined reference
'-DDEF=TEXT'	define a preprocessor replacement
'-Ipath'        additional include path for include files. Is evaluated before
                the XAINPUT environment variable. One path per '-I',
                multiple '-Ipath' allowed.
Omitting the errorfile or labelfile Parameter will cause xa to not write these files. Using '-x' will cause xa to take the name of the first source file and change the extension (on an Atari there is only one, like in DOS) to 'obj', 'err' and 'lab' respectively - if the old behaviour is selected with the '-x' option or the files are defined with "-l" and "-e". If no output file is given, "a.o65" is used.

Environment variables
You can use the variables XAOUTPUT and XAINPUT to adjust the directory structure. If source or include files are not found, the Path in XAINPUT is being searched for the files. The different paths are separated by a comma (','). XAINPUT gives the directory where the *.obj, *.err and *.lab files are put. If they are not set, there will be no search, respectively the files are saved to the current directory.

The label file is a readable ASCII-file and lists all the labels together with their block-count (see below) and their address. The error file lists the version of the assembler, date and time of the assembler run, all the error messages and the stuff being printed with #echo and #print and last but not least a statistics of used resources.

Some specific details
When using addressing modes that could be zeropage or absolute, zeropage will be taken if possible. This can be prevented by prefixing the address with a '!'. Then absolute addressing is taken, regardless of the address.

Values or Addresses can be expressed by arithmetik expressions with hierachy and bracket. The following operands are understood:

123       -decimal
$234      -hexadecimal
&123      -octal
%010110   -binary
*         -program counter
"A"       -ASCII-code
labelx    -label
-(lab1+1) -expression
The following operands can be used (third column is priority):

+         -addition                     9
-         -subtraction                  9
*         -multiplication               10
/         -integer-division             10
<<        -shift left                   8
>>        -shift right                  8
>=,=>     -more or equal                7 
<=,=<     -less or equal                7
<         -less                         7
>         -more                         7 
=         -equal                        6
<>,><     -not equal                    6
&&        -logical AND                  2
||        -Logical OR                   1
&         -Bitwise AND                  5
|         -Bitwise OR                   3
^         -Bitwise XOR                  4
Operators with higher priority are evaluated first. Brackets can be used as usual.

Valid expressions are, e.g.:

LDA       base+number*2,x
For Addressing modes that do not start with a bracket, you can even use a bracket at the beginning of an expression. Otherwise try this:

LDX       (1+2)*2,y                ; Wrong!
LDX       2*(1+2),y                ; Right!
Before an expression you can use these unitary operators:

<      Gives the low byte of the expression
>      Gives the high byte

LDA  #<adresse
Single Assembler statements are being separated by a ':' (You remember the C64 :-) or a newline. Behind Each statement, separated by a ';' you can write some comments. The next colon or a newline ends the comment and starts a new statement. In MASM compatibility mode ('-M' command line option), then a colon in a comment is ignored, i.e. the comment lasts till the newline.

Pseudo opcodes, Block structures and where Labels are valid

In addition to the 6502 opcodes you have the following Pseudo opcodes:
  • '.byt' and '.asc' are identical and save values to the memory (object file) bytewise.
    	.byt      value1,value2,value3, ...
    	.asc      "text1","text2", ...
  • '.word' does the same with words (2 Bytes).
    	.word     value1,value2, ...
  • '.dsb' fills a block with a given length with the value of fillbyte. If fillbyte is not given, zero is taken.
    	.dsb      length ,fillbte
  • '*=' changes the programm counter. The programm counter is not saved to the file as no rewind is being done. Just the internal counter is reloaded. If a value is given when the assembler has been started in relocation mode ('-R command line option), the assembler goes into no-relocation mode, i.e assembles everything without creating relocation table entries. For '*=' without a value, the assembler switches back to relocation mode. The absolute code is 'embedded' in the text segment, so the text segment program counter is increased by the length of the absolute code.

    One usage of this feature is for example to set the adress of buffers in BSS in the overlay memory on the oric:

    BigBuffer	.dsb 256

  • '.(' opens a new 'block'. All labels in a block are local, i.e. are only visible from inside the block - including sub-blocks. An error is returned if a label is defined that is already defined 'above'.
  • With '.)' the block is closed. You can have a stack of up to 16 blocks in each other (i.e. 16 times '.(' before the first '.)' will work, 17 not).
  • '.text', '.data', '.bss', '.zero' switch between the different segments.
    • The text segment is where the code goes in.
    • The data segment is where some initialized data goes in (it's actually like a second text segment).
    • The data segment might be allocated separated from the text segment.
    • The contents of the bss and the zero segment are not saved, just the labels are evaluated. Here goes the uninitialized data stuff.
    • The zero segment allows allocation of zeropage space, bss is normal address space.
    These opcodes can be used in relative and absolute mode.
  • '.align' aligns the current segment to a byte boundary given by the value. Allowed values are 2, 4, and 256. When using relative mode, the align value is written to the file header, such that relocation keeps the alignment.

    Note: It seems that this directive is totaly bugged, at least I never managed to ge it to do what I wanted to do. Since one of the most common use of allignement on the 6502 is to allign data on a multiple of 256 bytes (page), it's possible to hack around by using this work around:

    .dsb 256-(*&255)
    // The program counter from this point is alligned on the next multiple of 256
  • '.fopt' works like ".byte", but saves the bytes as a fileoption (see fileformat.txt). The length is computed automatically, so the first byte in the ".fopt" list of values should be the type. For example, the following line sets the filename for the object file.

    .fopt 0, "filename", 0


A label is defined by not being an opcode:

label1 LDA #0              ; assignes the programm counter
label2 =1234               ; explicit definition
label3 label4 label5       ; implicit programm counter
label6 label7 = 3	   ; label6 becomes the program counter, while
			   ; label7 is set to 3
label8:   sta label2	   ; As ':' divides opcodes, this is also
			   ; working
You can use more than one label for definition, except for explicit definition.
  • Labels are case sensitive.
  • If a label is proceeded by a '+', this label is defined global.
  • If a label is proceeded by a '&', this label is defined one level 'up' in the block hierachy, and you can use more than one '&'.
  • Redefinition of a label is possible by proceeding it with a dash '-':

    	-sysmem   +=4  ; here you can use ==, +=, -=, *=, /=, &=, |=
    	-syszp    =123


The preprocessor is very close to the one of the language C. So in addition to the ';'-comments you can also use C-like comments in '/*' and '*/'. Comments can be nested.

#include  "filename"     includes a file on exactly this position.
			 if the file is not found, it is searched using 

#echo  comment           gives a comment to the error file.

#file "filename"         Set the internal line counter in order to get correct line numbers at link phase

#print expression        prints an expression to the error file (after preprocessing and calculating)

#printdef DEFINED        prints the definition of a preprocessor define to the error file.

#define DEF  text	 defines 'DEF' by 'text'

#ifdef DEF               The source code from here to the following #endif
			 or #else is only assembled if 'DEF' is defined
			 with #define.

#else                    just else... (optionally)

#endif                   ends an #if-construct. This is a must to end #IF* 

#ifndef  DEF             .... if DEF is not defined

#if expression           .... if expression is not zero

#iflused label           .... if a label has already been used

#ifldef label            .... if a label is already defined
#iflused and #ifldef work an labels, not on preprocessor defs! With these commands a kind of library is easily built:

#iflused  label
#ifldef   label
#echo     label already defined, not from library
label     lda #0
You can have up to 15 #if* on stack before the first #endif

You can also use #define with functions, like in C.

#define mult(a,b)   ((a)*(b))
The preprocessor also allows continuation lines. I.e. lines that end with a '\' directly before the newline have the following line concatenated to it.

Known issues

Reported by Veltas the Wed 9th May 2018

Issue #22: .data section not working as expected
Details: If not specified, the .data section should by default start after the end of the .text section, they also should be saved in the binary, but apparently in non-relocatable mode this appears to be broken.

Reported by Christian the Thu 13th April 2017

Issue #19: MASM compatibility
Details: XA now handles correctly the ":" character in comments. The -M (MASM compatibility) option should be replaced by something that does nothing. (See: topic)

Resolved issues

Reported by Jede the Wed 12th April 2017
Fixed in OSDK 1.10

Issue #16: CC65 compatibility
Details: o65 object files generated by XA are not compatible with CC65 and require some binary patching. (See: topic)

Reported by Chema the Thu 23rd February 2017
Fixed in OSDK 1.9

Issue #12: Error with too many #defines
Details: The number of defines supported by XA is limited to 2340, expanding the value to something significantly larger would be welcome.

Reported by Jede the Mon 13th February 2017
Fixed in OSDK 1.10

Issue #14: Need user definable XA parameters
Details: Add a new parameter,OSDKXAPARAMS, which can be used to pass additional parameters to the assembler. A typical usage would be to pass additional defines using the -D notation.

Reported by Godzil the Mon 1st June 2015
Fixed in OSDK 1.8

Issue #9: Restrict available 6502 instructions
Details: At the moment XA allows any 6502 instructions, such as PLX, PLY, PHX, PHY which are not available in the NMOS version of the 6502 used on the Oric. Ideally they should be refused by default, except if allowed from a command line flag.

comments powered by Disqus
Coverity Scan Build Status
Version 2.2.5
  • Added a new '-cc' command line parameter to allow compatibility with the CC65 toolchain
Version 2.2.4
  • Extended the values for a number of hardcoded defines (labels, blocks, number of open files, etc...)
Version 2.2.3
  • Fixed a stupid crash happening when no filename was provided for the symbols or error files.
Version 2.2.2
  • Normally C, C++ and asm comments should be handled correctly, and ignored if in a quoted string.
Version 2.2.1
  • hopefully fixed a problem of data corruption happening when a line parsed during pass1 generated more than 127 bytes of tokens for the pass2 to decode.
Version 2.1.7
  • removed the '-x' switch from XA code. I will progressively remove all the legacy stuff when replacement features already exist, in this case -o, -e and -l
Version 2.1.6
  • debogged the #file directive that was actually giving valid file names only for the errors that happened during pass 1.
Version 2.1.5
  • added a #file directive
  • corrected some error messages
Version 2.1.4h (25nov1998)
  • preprocessor fix, improve file65.
Version 2.1.4g (25nov1998)
  • fix '!' addressing syntax, add reference list for labels.
Version 2.1.4f (21apr1998)
  • cross-compilable for DOS with make dos
Version 2.1.4e
  • some bugfixes concerning o65 fileformat, as well as a fileformat change.
Version 2.1.4
  • preprocessor understands continuation lines, more command line options available.
Version 2.1.3
  • Segments can be aligned to 2, 4, or 256 byte address boundary.
    A new pseudo-opcode has been introduced, '.align', that aligns the address in the current segment to the value given.
Version 2.1.2a
  • linker, relocator and file utility for 'o65' format included, some bugs fixed.
Version 2.1.1e
  • started supporting the 'o65' file format
Version 2.0.7d
  • now understands the #undef preprocessor directive.
Version Version 2.0.7.c
  • the include statements can now contain '\' or '/' as directory separator, and both are accepted (with DIRCHAR='/').