Free SNES assembler

This program reads symbolic 65816 or 65c816 machine code and compiles (assembles) it into a relocatable object file.

The produced object file is binary-compatible with those made with XA65.

This program was born when Bisqwit needed a relocatable object -producing snes assembler and XA65 had too many bugs in it.
He was unable to fix the XA65 source so he started his own project, aiming for enough compatibility to be able to use the assembly files already written in the XA65 syntax.
Since that, this program has been growing to meet the needs in SNES game and patch development.

This package also contains a linker.
See http://bisqwit.iki.fi/src/snescom-demo/ - there's a demo explaining how this all works.

The following mnemonics are supported:

adc, and, asl, bcc, bcs, beq, bit, bmi, bne, bpl, bra, brk, brl, bvc, bvs, clc, cld, cli, clv, cmp, cop, cpx, cpy, db , dec, dex, dey, eor, inc, inx, iny, jml, jmp, jsl, jsr, lda, ldx, ldy, lsr, mvn, mvp, nop, ora, pea, pei, per, pha, phb, phd, phk, php, phx, phy, pla, plb, pld, plp, plx, ply, rep, rol, ror, rti, rtl, rts, sbc, sec, sed, sei, sep, sta, stp, stx, sty, stz, tax, tay, tcd, tcs, tdc, trb, tsb, tsc, tsx, txa, txs, txy, tya, tyx, wai, xba, xce

All the standard addressing modes of the 65816 cpu are supported.

Syntax:

• Implied: nop; clc
• Immediate: lda #value; rep #value etc (size may depend on an operand size setting)
• Short relative: bra end
• Long relative: brl end; per end+2
• Direct: lda $12
• Direct indexed: lda $12,x; lda $12,y
• Direct indirect: lda ($12); pei ($12)
• Direct indexed indirect: lda ($12,x)
• Direct indirect indexed: lda ($12),y
• Direct indirect long: lda [$12]
• Direct indirect indexed long: lda [$12],y
• Absolute: lda $1234
• Absolute indexed: lda $1234,x; lda $1234,y
• Absolute long: lda $123456
• Absolute indexed long: lda $123456,x
• Stack-relative: lda $12,s
• Stack-relative indirect indexed: lda ($12,s),y
• Absolute indirect: lda ($1234)
• Absolute indirect long: lda [$1234]
• Absolute indexed indirect: lda ($1234,x)
• MVN/MVP: mvn $7E,$7F

For reference, in Intel syntax it would be something like this (not supported by snescom):

• Implied: nop; clc
• Immediate: lda value; rep value etc (size may depend on an operand size setting)
• Short relative: bra end
• Long relative: brl end; per end+2
• Direct: lda [$00:d+$12]
• Direct indexed: lda [$00:d+$12+x]; lda [$00:d+$12+y]
• Direct indirect: lda [db:[$00:d+$12]]; pei [db:[$00:d+$12]]
• Direct indexed indirect: lda [db:[$00:d+$12+x]]
• Direct indirect indexed: lda [db:[$00:d+$12]+y]
• Direct indirect long: lda [long[$00:d+$12]]
• Direct indirect indexed long: lda [long[$00:d+$12]+y]
• Absolute: lda [db:$1234]
• Absolute indexed: lda [db:$1234+x]; lda [db:$1234+y]
• Absolute long: lda [$12:$3456]
• Absolute indexed long: lda [$12:$3456+x]
• Stack-relative: lda [$00:$12+s]
• Stack-relative indirect indexed: lda [db:[$00:$12+s]+y]
• Absolute indirect: lda [db:[$1234]]
• Absolute indirect long: lda [long[db:$1234]]
• Absolute indexed indirect: lda [db:[db:$1234+x]]
• MVN/MVP: mvn $7E,$7F

The pseudo ops .as, .al, .xs & .xl are used to decide what size accumulator and index mode for the assembler to use. .as and .xs are for 8bit operands, and .al and .xl are for 16 bit operands.

I've found it handy to define these macroes:

#define SET_8_BIT_A()   sep #$20 : .as
#define SET_16_BIT_A()  rep #$20 : .al
#define SET_8_BIT_X()   sep #$10 : .xs
#define SET_16_BIT_X()  rep #$10 : .xl

#define SET_8_BIT_AX()  sep #$30 : .xs : .as
#define SET_16_BIT_AX() rep #$30 : .xl : .al

In addition to these modes, there are several operand prefixes that can be used to force a certain operand size/type.

• lda !$f0 would use 16-bit address instead of direct page.
• lda @$1234 would use a long (24-bit) address instead of absolute 16-bit address.
• lda #<var can be used to load the lower 8 bits of an external variable.
• lda #>var can be used to load the upper 8 bits of an external variable.
• lda #^var can be used to load the segment part (8 bits) of an external variable.
• lda @var,x can be used to use absolute-indexed-long mode instead of the default, absolute.

Expressions are supported. These are valid code:

• bra somewhere+1
• lda #!address + $100
• ldy #$1234 + ($6C * 3)

Code, labels and data can be generated to four segments: text, data, zero and bss.
Use .text, .data, .zero and .bss respectively to select the segment.
However, only the contents of text and data segments are saved into the o65 file. Labels are saved in all segments.

Comments begin with a semicolon (;) and end with a newline.
A colon is allowed to appear in comment.

Commands are separated by newlines and colons (:).

You can use a command like *= $34F200 to change where the code goes by default.
With IPS this is especially useful.
You can change the code pointer as many times as you wish, but unless you're generating an IPS file, all code must be a continuous block.

The label - can be defined for branches backward and + for branches forward.
Example:

        ; Space-fill the buffer to end
        phx
         cpx #$0010
         bcs +     ;jumps to the next "+"
         SET_8_BIT_A()
         lda #$FF
-        sta $94A0,x
         sta $94B0,x
         inx
         cpx #$0010
         bcc -     ;jumps to the previous "-"
         SET_16_BIT_A()
+        lda W_VRAMADDR
         sta @$002116
        pla

snescom uses GCC as a preprocessor.
You can use #ifdef, #ifndef, #define, #if, #endif and #include like in any C program. (See bugs)

Currently snescom only produces relocatable object files (O65) or non-relocatable patch files (IPS).
The O65 file format has been documented by André Fachat for the XA65 project.

If --jumps option is enabled, all short jumps are automatically turned into near jumps when needed. (See bugs)

This version of snescom allows you to create IPS files.
This IPS format has been extended to allow you to specify global symbols and externs to be patched later.
In the generated format:

• If the patch address is $000001, it means this entry is a dynamic symbol. What follows is an asciiz string (the symbol name), the address (3-byte lsb-first integer) and the relocation size in bytes (1 byte).
• If the patch address is $000002, it means this entry is a global symbol. What follows is an asciiz string (the symbol name) and the address (3-byte lsb-first integer) where this symbol is located in.

If you only feed the IPS file to sneslink, you do not need to know about this extension.

By default, O65 objects are linked to any free location in the ROM.
IPS files are linked to predefined locations.

With the .link statement, you can change that.

.link page $3F declares that this object should be placed into page $3F.

.link group 1 declares that this object should be placed to the same page together with all other objects that want to be linked in group 1. This is useful when you want to ensure that certain tables or routines go to the same page, even if they are not in the same compilation unit.
The actual page is determined during link time, and you can get the page by using 24-bit (@) or segment reference (^) to a symbol from those modules.

The linkage selection statement is only honored by the sneslink program, distributed in this archive.

sneslink has currently been tested only with HiROMs.

In the source distribution there are some C++ modules that can be used to handle the o65 files.
There are also some example assembler files (copied from Chronotools).

There's also an example program demonstrating the linker at http://bisqwit.iki.fi/src/snescom-demo/.

Sep 23 2003; 0.0.0 started working with the project.
Sep 28 2003; 1.0.0 initial release. All features working.
Sep 29 2003; 1.1.0 added feature: branch labels.
Sep 29 2003; 1.2.0 bugfix: didn't resolve correctly when different scopes had a label with the same name; added warning options
Oct 3 2003; 1.3.0 added feature: jump optimization.
Oct 8 2003; 1.3.1 bugfix: jump optimization only supports two passes. Requiring more now produces an error.
Oct 13 2003; 1.3.2 bugfix: accidentally had broke MVN/MVP some time ago
Nov 3 2003; 1.3.2.1 bugfix: undef; broken HTML in README.html
Nov 3 2003; 1.3.3 added some alternative methods (pea #$8000, mvn $7E7F)
Dec 30 2003; 1.3.4 bugfix; error-exit now deletes the broken output file
Jan 2 2004; 1.3.4.1 bugfix; added missing files
Jan 23 2004; 1.3.4.2 handles memory addressing a bit differently internally
Jan 24 2004; 1.4.0 adds IPS output support
Jan 24 2004; 1.4.1 added new program: IPS and O65 linker (BETA!)
Jan 24 2004; 1.4.1.1 added feature: string constants in .byt
Jan 28 2004; 1.4.1.2 IPS bugfixes; linker improved
Jan 31 2004; 1.4.1.3 some internal changes
Jan 31 2004; 1.4.1.4 more internal changes - aiming towards multisegment linking
Feb 3 2004; 1.4.2 improved expression parsing
Feb 22 2004; 1.5.0 added the linkage selection statement
Jan 30 2005; 1.5.0.1 adds a very hackish kludge to avoid three-pass jump fixing in a certain case
Nov 21 2005; 1.5.1 symbol support in disassembler (when given an O65 or IPS file)

• #included files aren't being properly preprocessed.
• The --jumps option is not failsafe - three pass jump optimization is not supported.

snescom has been written by Joel Yliluoma, a.k.a. Bisqwit,
and is distributed under the terms of the General Public License (GPL).

If you happen to see this program useful for you, I'd appreciate if you tell me :) Perhaps it would motivate me to enhance the program.

The official home page of snescom is at http://iki.fi/bisqwit/source/snescom.html.
Check there for new versions.