# TMS34010 Disassembler Project

This is a (hopefully portable) TMS34010 disassembler which has built-in options to 'intelligently' analyze ROM contents and determine which areas are most likely to be executed, and then disassembles the corresponding areas as code or data. Also include are a couple of options to load multiple ROMs in 16- and 8-bit format. The skeleton with the code base and heuristics was human-written, but I had Claude Code fill in the gaps and test it extensively.

The output format can be assembled with as34, which is a TMS34010 assembler that (as far as I remember) I got from a Texas Instruments' SDK/Devkit in the early 2000's.

---

## dis34 Usage

```
dis34 [-a] [-b base_addr] [-s skip] [file]
dis34 [-a] -e file1 [file2 ...]
dis34 [-a] -E lo1 hi1 [lo2 hi2 ...]
  -a  analyze mode: heuristic code/data separation
  -b  base bit-address (hex, default 0)
  -e  EPROM mode: concatenate 16-bit files, last byte at top of address space
  -E  8-bit EPROM mode: interleave pairs (lo,hi), last byte at top
  -s  skip bytes at start of file
  -p  import code entry points from file (one hex address per line)
  -P  export discovered entry points to file
  -l  show address prefix on each output line (off by default)
  -o  show opcode bytes as comment after each instruction
  -h  show help
  reads stdin if no file given
```

Examples:
```
./dis34 -s 80 -b fff00000 somefile.bin
./dis34 -a -E u4.bin u7.bin                          # single pair
./dis34 -a -E u4.bin u7.bin u6.bin u9.bin             # multiple pairs
./dis34 -a -P points.txt -E u4.bin u7.bin             # export entry points
./dis34 -a -p points.txt -E u4.bin u7.bin             # import entry points
```

---

## TMS34010 Instruction Set Architecture

### Registers

- **A-file**: a0-a14, accessed when register bit 4 = 0
- **B-file**: b0-b14, accessed when register bit 4 = 1
- **sp**: register 15 in either file (stack pointer), encoded as `0x0F`
  (A15, printed `sp`) or `0x1F` (B15, printed `b15`).  Both are the same
  physical stack pointer, but the encoding names a different file; a
  disassembler must not collapse them.
- Register encoding: 5-bit field, bit 4 = file select, bits 3:0 = register number

### Instruction Word Format

All instructions are 16-bit words (little-endian). Multi-word instructions
follow with 16-bit or 32-bit immediates/addresses.

```
Bits 15:9  - Primary opcode (op & 0xFE00)
Bits  8:5  - Sub-opcode or source register (RS)
Bits  4:0  - Destination register (RD), bit 4 = file select
Bit   9    - Field selector for MOVE/SETF (0 or 1)
```

### Primary Decode: `op & 0xFE00`

| Range | Instructions |
|-------|-------------|
| 0x0000 | rev, emu, exgpc, getpc, jump, getst, putst, popst, pushst |
| 0x0200 | nop, clrc, movb @,@, dint, abs, neg, negb, not |
| 0x0400-0x0600 | sext, zext, setf, move r/@, move @/r, move @/@, movb r/@, move @/r (with field) |
| 0x0800 | trap, call, reti, rets, mmtm, mmfm, moviw, movi (32-bit) |
| 0x0A00 | addiw, addi, cmpiw, cmpi, andi, ori, xori, subiw |
| 0x0C00 | subi (32-bit), callr, calla, eint, dsj, dsjeq, dsjne, setc |
| 0x0E00 | pixblt l/xy/b variants, fill l/xy |
| 0x1000-0x1200 | addk / inc (k=1) |
| 0x1400-0x1600 | subk / dec (k=1) |
| 0x1800-0x1A00 | movk |
| 0x1C00-0x1E00 | btst (K-form, 1's complement) |
| 0x2000-0x2200 | sla (K-form) |
| 0x2400-0x2600 | sll (K-form) |
| 0x2800-0x2A00 | sra (K-form, 2's complement: 32-stored) |
| 0x2C00-0x2E00 | srl (K-form, 2's complement: 32-stored) |
| 0x3000-0x3200 | rl (K-form) |
| 0x3800-0x3E00 | dsjs (5-bit displacement, bit 10 = negate) |
| 0x4000-0x6E00 | Register-register ALU (add, sub, cmp, and, or, xor, shift, etc.) |
| 0x8000-0xBC00 | MOVE/MOVB addressing modes |
| 0xC000-0xCE00 | Conditional jumps (JR/JA) |
| 0xD000-0xD600 | MOVE *rs(d),*rd+; MOVE @abs,*rd+; exgf |
| 0xDE00 | LINE 0 / LINE 1 |
| 0xE000-0xEE00 | addxy, subxy, cmpxy, cpw, cvxyl, movx, movy |
| 0xF000-0xFC00 | PIXT / drav |

### Complete Opcode Table

#### Register-Register (REGREG): `opval | (rs<<5) | rd`

| Mnemonic | Opval | Notes |
|----------|-------|-------|
| add | 0x4000 | |
| addc | 0x4200 | |
| sub | 0x4400 | |
| subb | 0x4600 | |
| cmp | 0x4800 | |
| btst | 0x4A00 | Register form |
| and | 0x5000 | |
| andn | 0x5200 | |
| or | 0x5400 | |
| xor | 0x5600 | clr when rs==rd |
| divs | 0x5800 | |
| divu | 0x5A00 | |
| mpys | 0x5C00 | |
| mpyu | 0x5E00 | |
| sla | 0x6000 | Register form |
| sll | 0x6200 | Register form |
| sra | 0x6400 | Register form |
| srl | 0x6600 | Register form |
| rl | 0x6800 | Register form |
| lmo | 0x6A00 | |
| mods | 0x6C00 | |
| modu | 0x6E00 | |
| addxy | 0xE000 | |
| subxy | 0xE200 | |
| cmpxy | 0xE400 | |
| cpw | 0xE600 | |
| cvxyl | 0xE800 | |
| movx | 0xEC00 | |
| movy | 0xEE00 | |
| drav | 0xF600 | |

#### One-Register (ONEREG): `opval | rd`

| Mnemonic | Opval |
|----------|-------|
| rev | 0x0020 |
| exgpc | 0x0120 |
| getpc | 0x0140 |
| jump | 0x0160 |
| getst | 0x0180 |
| putst | 0x01A0 |
| abs | 0x0380 |
| neg | 0x03A0 |
| negb | 0x03C0 |
| not | 0x03E0 |
| call | 0x0920 |
| inc | 0x1020 |
| dec | 0x1420 |

#### Implied (no operands)

| Mnemonic | Opval |
|----------|-------|
| emu | 0x0100 |
| popst | 0x01C0 |
| pushst | 0x01E0 |
| nop | 0x0300 |
| clrc | 0x0320 |
| dint | 0x0360 |
| reti | 0x0940 |
| eint | 0x0D60 |
| setc | 0x0DE0 |
| pixblt l,l | 0x0F00 |
| pixblt l,xy | 0x0F20 |
| pixblt xy,l | 0x0F40 |
| pixblt xy,xy | 0x0F60 |
| pixblt b,l | 0x0F80 |
| pixblt b,xy | 0x0FA0 |
| fill l | 0x0FC0 |
| fill xy | 0x0FE0 |
| line 0 | 0xDF1A |
| line 1 | 0xDF9A |

#### Immediate-Long + Register (ILREG): `opval | rd`, then 32-bit immediate

| Mnemonic | Opval | Notes |
|----------|-------|-------|
| addi | 0x0B20 | Also word form at 0x0B00 |
| movi | 0x09E0 | Also word form at 0x09C0 |
| ori | 0x0BA0 | |
| xori | 0x0BC0 | |

#### Negated Immediate-Long + Register (NILREG): stored as ~value

| Mnemonic | Opval | Notes |
|----------|-------|-------|
| andi | 0x0B80 | 1's complement |
| cmpi | 0x0B60 | 1's complement; also word form at 0x0B40 |
| subi | 0x0D00 | 1's complement; also word form at 0x0BE0 |

#### Constant + Register (KREG): `opval | (k<<5) | rd`, K range 1-32 (0=32)

| Mnemonic | Opval | Notes |
|----------|-------|-------|
| addk | 0x1000 | inc when k=1 |
| subk | 0x1400 | dec when k=1 |
| movk | 0x1800 | |

#### Constant/Register dual-form (KREGORREGREG)

| Mnemonic | K-form | R-form | K encoding |
|----------|--------|--------|------------|
| sla | 0x2000 | 0x6000 | Direct |
| sll | 0x2400 | 0x6200 | Direct |
| rl | 0x3000 | 0x6800 | Direct |

#### Negated Constant/Register dual-form (NKREGORREGREG)

| Mnemonic | K-form | R-form | K encoding |
|----------|--------|--------|------------|
| btst | 0x1C00 | 0x4A00 | 1's complement: (~op>>5)&0x1F |
| sra | 0x2800 | 0x6400 | 2's complement: 32-((op>>5)&0x1F) |
| srl | 0x2C00 | 0x6600 | 2's complement: 32-((op>>5)&0x1F) |

#### Special Instructions

| Mnemonic | Opval | Format |
|----------|-------|--------|
| callr | 0x0D3F | Fixed opcode + 16-bit PC-relative displacement |
| calla | 0x0D5F | Fixed opcode + 32-bit absolute address |
| mmtm | 0x0980 | opval\|rd + 16-bit register bitmask (MSB=reg0) |
| mmfm | 0x09A0 | opval\|rd + 16-bit register bitmask (LSB=reg0) |
| trap | 0x0900 | opval\|(trap_num & 0x1F) |
| rets | 0x0960 | opval\|(pop_count & 0x1F) |
| setf | 0x0540/0x0740 | opval\|fs\|(fe<<5), field in bit 9 |
| exgf | 0xD500 | opval\|rd, field in bit 9 |
| dsj | 0x0D80 | opval\|rd + 16-bit displacement |
| dsjeq | 0x0DA0 | opval\|rd + 16-bit displacement |
| dsjne | 0x0DC0 | opval\|rd + 16-bit displacement |
| dsjs | 0x3800+ | opval\|rd\|(disp<<5), bit 10 = negate |

### MOVE Addressing Modes (18 variants)

| Opval | Format | Field |
|-------|--------|-------|
| 0x4C00 | move rs, rd | No |
| 0x4E00 | move rs, rd (cross-file) | No |
| 0x8000/0x8200 | move rs, *rd | Yes |
| 0x8400/0x8600 | move *rs, rd | Yes |
| 0x8800/0x8A00 | move *rs, *rd | Yes |
| 0x9000/0x9200 | move rs, *rd+ | Yes |
| 0x9400/0x9600 | move *rs+, rd | Yes |
| 0x9800/0x9A00 | move *rs+, *rd+ | Yes |
| 0xA000/0xA200 | move rs, -*rd | Yes |
| 0xA400/0xA600 | move -*rs, rd | Yes |
| 0xA800/0xAA00 | move -*rs, -*rd | Yes |
| 0xB000/0xB200 | move rs, *rd(disp) | Yes |
| 0xB400/0xB600 | move *rs(disp), rd | Yes |
| 0xB800/0xBA00 | move *rs(d), *rd(d) | Yes |
| 0xD000/0xD200 | move *rs(d), *rd+ | Yes |
| 0x0580/0x0780 | move rd, @addr | Yes |
| 0x05A0/0x07A0 | move @addr, rd | Yes |
| 0x05C0/0x07C0 | move @addr1, @addr2 | Yes |
| 0xD400/0xD600 sub0 | move @addr, *rd+ | Yes |

"Field" column: `Yes` = appends `, 0` or `, 1` to select field register.
Paired opcodes differ by bit 9 (field select): low = field 0, high = field 1.

### MOVB Addressing Modes (9 variants)

| Opval | Format |
|-------|--------|
| 0x8C00 | movb rs, *rd |
| 0x8E00 | movb *rs, rd |
| 0x9C00 | movb *rs, *rd |
| 0xAC00 | movb rs, *rd(disp) |
| 0xAE00 | movb *rs(disp), rd |
| 0xBC00 | movb *rs(d), *rd(d) |
| 0x0340 | movb @addr1, @addr2 |
| 0x05E0 | movb rd, @addr |
| 0x07E0 | movb @addr, rd |

### PIXT Addressing Modes (6 variants)

| Opval | Format |
|-------|--------|
| 0xF000 | pixt rs, *rd, xy |
| 0xF200 | pixt *rs, xy, rd |
| 0xF400 | pixt *rs, xy, *rd, xy |
| 0xF800 | pixt rs, *rd |
| 0xFA00 | pixt *rs, rd |
| 0xFC00 | pixt *rs, *rd |

### Conditional Jump Encoding

Condition code in bits 11:8 of the instruction word:

| CC | Code | Meaning |
|----|------|---------|
| 0 | uc | Unconditional |
| 1 | p | Plus (positive) |
| 2 | ls | Low or same |
| 3 | hi | High |
| 4 | lt | Less than |
| 5 | ge | Greater or equal |
| 6 | le | Less or equal |
| 7 | gt | Greater than |
| 8 | c | Carry |
| 9 | nc | No carry |
| 10 | eq | Equal (zero) |
| 11 | ne | Not equal |
| 12 | v | Overflow |
| 13 | nv | No overflow |
| 14 | n | Negative |
| 15 | nn | Not negative |

Jump forms:
- **JR cc** (relative, 2 words): `0xC000 | (cc<<8) | 0x00` + 16-bit displacement
- **JR cc** (short, 2 words): `0xC000 | (cc<<8) | disp8` (8-bit signed displacement)
- **JA cc** (absolute, 3 words): `0xC000 | (cc<<8) | 0x80` + 32-bit address

Only low byte `0x80` is the absolute form.  Low byte `0x00` carries a 16-bit
*displacement* and is still `JRcc`; the word count must not be used to invent
an absolute mnemonic for it.

Displacement formula:
- 16-bit: `target = pc + 32 + (disp << 4)`
- 8-bit: `target = pc + 16 + (disp << 4)`

### The `issrl` Flag and NKREGORREGREG K-Encoding

The NKREGORREGREG instruction class (BTST, SRA, SRL) uses negated K constants,
but the negation method differs between instructions, controlled by the `issrl`
flag in the as34 assembler:

| Instruction | `issrl` | Negation | Encode formula | Decode formula |
|-------------|---------|----------|----------------|----------------|
| BTST | 0 | 1's complement | `(~K) & 0x1F` | `(~stored) & 0x1F` |
| SRA | 1 | 2's complement | `(-K) & 0x1F` | `32 - stored` |
| SRL | 1 | 2's complement | `(-K) & 0x1F` | `32 - stored` |

The two methods produce different stored values for every K from 1-31:

```
K:   1   2   3  ...  27  28  29  30  31
1's: 30  29  28 ...   4   3   2   1   0
2's: 31  30  29 ...   5   4   3   2   1
```

Using the wrong decode gives K±1 errors. In `dis34.c`:
- BTST (0x1C00): `(~op >> 5) & 0x1f` — 1's complement — correct
- SRA (0x2800): `32 - ((op >> 5) & 0x1f)` — 2's complement — correct
- SRL (0x2C00): `32 - ((op >> 5) & 0x1f)` — 2's complement — correct


---

### Round-Trip Design Decisions

1. **`jrcc` for both relative forms** -- the 8-bit and 16-bit displacement
   forms are both `JRcc` (2 words); only the 3-word `0x80` form is `JAcc`.
   (Until 2026-10-18 the 16-bit form was printed as `jacc`, which made a
   linear disassembly's control flow wrong and forced the round-trip
   converter to rewrite it back to `jrcc`.)

2. **Branch-only labels** -- Only branch/call targets (jr, ja, callr, calla,
   dsj, dsjeq, dsjne, dsjs) get labels. MOVI and other data-reference
   instructions keep literal hex addresses to avoid address shifts when
   code section size changes.

3. **Raw data section** -- dis2asm.py reads the data section directly from the
   binary (-i flag) rather than from dis34 output, because the disassembler
   may misinterpret data words as instructions.

### Known Formatting Differences from tos.asm

These are formatting-only differences that do not affect binary output:

1. **`move a0, a1` vs `move a0, a1, 1`** -- Field specifier not encoded in
   reg-to-reg MOVE; assembler accepts but ignores it
2. **Hex vs decimal** -- dis34 always uses `0x` prefix; tos.asm mixes decimal
   and hex
3. **MMTM/MMFM register order** -- dis34 outputs registers in bit order
   (ascending); assembler accepts any order
4. **Data section** -- Disassembles as `.word` directives and garbage
   instructions (expected for a linear disassembler)

---

## Analyze Mode (`-a`)

Heuristic code/data separation using two passes:

1. **Pass 1 (sequential scan)**: Finds blocks of valid instructions terminated
   by unconditional transfers, validates them (min 4 instructions, has a branch,
   consistent register file usage, valid branch targets).

2. **Pass 2 (recursive descent)**: Seeds worklist from pass 1 branch targets
   and TMS34010 hardware vector table (traps 0-31 at 0xFFFFFFE0 downward).
   Traces execution paths, validating each block (min 2 instructions, register
   file consistency) before committing.

### Unified Address Space in EPROM Mode

When multiple EPROM pairs are loaded (`-E lo1 hi1 lo2 hi2 ...`), the entire
combined ROM is analyzed as a single unified address space. The last pair maps
to the top of the 32-bit address space, where the TMS34010 hardware vector
table lives (0xFFFFFFE0 downward). Branches between pairs are followed
normally, allowing the recursive descent to trace execution across the full ROM.

### Heuristic Filters

- **Register file consistency**: Real TMS34010 code uses either A-file or B-file
  registers consistently. Blocks with >=25% minority register file usage are
  rejected as likely data.
- **Minimum trace length**: Single-instruction traces from recursive descent are
  rejected to prevent cascading false positives.

### Entry Point Import/Export (`-p` / `-P`)

The `-P FILE` option exports all discovered code entry points (branch/call
targets and vector table handlers) to a file, one hex address per line. This
allows inspection and manual curation of the entry point list.

The `-p FILE` option imports entry points from a file and uses them as
additional seeds for the recursive descent analysis. The file format is one
hex address per line; lines starting with `#` are comments, blank lines are
ignored.

Typical workflow:
1. Run with `-P points.txt` to export discovered entry points
2. Edit `points.txt` to add known code addresses or remove false positives
3. Re-run with `-p points.txt` to improve analysis coverage

---

## Labels and Vector Table

### Labels

Branch and call targets are automatically labelled using the format
`L` followed by the uppercase hexadecimal bit-address (e.g., `LFFF00820:`).
Only addresses that are actually referenced by a branch, call, or vector table
entry receive labels. The corresponding branch/call instructions use the label
name instead of a raw hex address.

### Vector Table Display

When the ROM maps to the top of the 32-bit address space (as in EPROM modes),
the TMS34010 hardware vector table (traps 0–31 at 0xFFFFFFE0 downward) is
displayed as labeled `.long` entries with standard vector names:

```
RESET:
ffffffe0:       .long  LFFCEF750
INT1:
ffffffc0:       .long  LFFD7D290
INT2:
ffffffa0:       .long  LFFD7DDB0
ILLOP:
fffffe80:       .long  LFF8632E0
```

The handler addresses are shown as label references, and the corresponding
handler code receives matching labels in the disassembly output.

---

## TMS34010 I/O Registers

The TMS34010 has 28 memory-mapped I/O registers in the bit-address range
`0xC0000000`–`0xC00001F0`, spaced 0x10 apart (16 bits each). The disassembler
automatically annotates instructions that reference these addresses with the
register name as a comment.

| Bit Address    | Register | Description                     |
|----------------|----------|---------------------------------|
| `0xC0000000`   | HESYNC   | Horizontal End Sync             |
| `0xC0000010`   | HEBLNK   | Horizontal End Blank            |
| `0xC0000020`   | HSBLNK   | Horizontal Start Blank          |
| `0xC0000030`   | HTOTAL   | Horizontal Total                |
| `0xC0000040`   | VESYNC   | Vertical End Sync               |
| `0xC0000050`   | VEBLNK   | Vertical End Blank              |
| `0xC0000060`   | VSBLNK   | Vertical Start Blank            |
| `0xC0000070`   | VTOTAL   | Vertical Total                  |
| `0xC0000080`   | DPYCTL   | Display Control                 |
| `0xC0000090`   | DPYSTRT  | Display Start Address           |
| `0xC00000A0`   | DPYINT   | Display Interrupt               |
| `0xC00000B0`   | CONTROL  | Control Register                |
| `0xC00000C0`   | HSTDATA  | Host Data                       |
| `0xC00000D0`   | HSTADRL  | Host Address Low                |
| `0xC00000E0`   | HSTADRH  | Host Address High               |
| `0xC00000F0`   | HSTCTLL  | Host Control Low                |
| `0xC0000100`   | HSTCTLH  | Host Control High               |
| `0xC0000110`   | INTENB   | Interrupt Enable                |
| `0xC0000120`   | INTPEND  | Interrupt Pending               |
| `0xC0000130`   | CONVSP   | Conversion Source Pitch         |
| `0xC0000140`   | CONVDP   | Conversion Destination Pitch    |
| `0xC0000150`   | PSIZE    | Pixel Size                      |
| `0xC0000160`   | PMASK    | Pixel/Plane Mask                |
| `0xC00001B0`   | DPYTAP   | Display Tap Point               |
| `0xC00001C0`   | HCOUNT   | Horizontal Counter (read-only)  |
| `0xC00001D0`   | VCOUNT   | Vertical Counter (read-only)    |
| `0xC00001E0`   | DPYADR   | Display Address                 |
| `0xC00001F0`   | REFCNT   | DRAM Refresh Counter            |

Registers at `0xC0000170`–`0xC00001A0` are reserved/unused.

References to these registers also serve as a heuristic signal in analyze mode:
blocks that access I/O registers are almost certainly real code, so the
validation requirements are relaxed for such blocks.