A fast Mandelbrot generator for 8-bit computers, written in 100% assembly.
This is a fork of the original
mandelbr8 by 0x444454.
See Attribution.
- 40-pixel stripe Mariani-Silver rendering on C64: 27 % faster multicolor, 33 % faster contour.
- 320×200 monochrome contour mode for C64 (separate binary).
- Edge-triggered input: one press, one action. Pan and zoom work reliably under VICE warp mode.
- Pan and zoom step sizes raised to ~4 % and ~8 % per press.
- Vertical centering fixes for C64, C128 VIC-IIe, and C128 VDC.
- Luma-ordered palettes for C64, C128, TED, and VIC-20.
- Download a binary from the
release/folder for your machine. - Load and run it the usual way:
LOAD"*",8,1 RUN - Use a joystick to pan and zoom. See Controls.
- In VICE, turn on warp mode to shorten render times.
| Target | Notes |
|---|---|
| Commodore 64 | Multicolor and contour modes; optional Kawari acceleration |
| Commodore 128 | VIC-IIe (40-col) and VDC (80-col) modes |
| Plus/4 / C16 | TED chip; 64 KB required |
| VIC-20 | 16 KB memory expansion required (Block 1 + Block 2 minimum) |
| PET | 8 KB+ RAM; optional color support |
| CBM-II (B128) | 6xx and 7xx series |
| Atari XL/XE | 64 KB; uses GTIA 16-luma mode |
| BBC Micro B | 32 KB |
| Target | Joystick port / input |
|---|---|
| C64 / C128 | Port 2 |
| Plus/4 / C16 | Port 1 |
| VIC-20 | Port 1 |
| Atari XL/XE | Port 1 |
| PET / CBM-II | WASD for directions, SHIFT for fire |
| BBC Micro | Analog port 1, or cursor keys + Shift |
| Input | Effect |
|---|---|
| Direction (no fire) | Pan view (~4 % per press) |
| Fire + Up | Zoom in (~8 % per press) |
| Fire + Down | Zoom out |
| Fire + Right | Increment max iterations (cap 255) |
| Fire + Left | Decrement max iterations (floor 2) |
Each press triggers one action. Zoom saturates at the precision floor, about 12× from the default view.
Default max_iter = 32 gives 16 contour bands.
PET has no bitmap mode; the renderer uses PETSCII glyphs for shading.
CBM-II uses an 80-column text-mode render with PETSCII shading.
A SuperCPU lifts the per-iteration cost enough that very deep zooms
become practical. With max_iter raised via Fire+Right, more iter
bands appear and fine fractal detail at the precision floor stays
visible.
| Multicolor | Contour |
|---|---|
 |
 |
| Target | Lo-res preview | Hi-res render |
|---|---|---|
| C64 / C128 40-col / Plus/4 | 40×25, 16 colors | 160×200, 16 colors (multicolor bitmap) |
| C64 contour | (skipped) | 320×200 monochrome |
| C128 80-col (VDC) | 80×50, 16 colors | 160×100, 16 colors (64 KB VDC RAM required) |
| VIC-20 | 22×22, 16 colors | 88×176, 4 colors (multicolor bitmap) |
| PET | 40×25 mono / 16-color | not supported |
| CBM-II | 80×25 mono | not supported |
| Atari XL/XE | 40×25, 16 grey shades | 80×200, 16 grey shades (ANTIC F.1) |
| BBC Micro | 40×32, 8 colors | 160×256, 8 colors |
Turn on warp mode to shorten render times. VICE emulates faster than real time without changing the result, and the input handler stays accurate.
VICE shortcut: Alt+W (macOS) or Page Up (Windows / Linux). Or Speed → Warp mode in the menu.
Built with 64TASS (≥ 1.59).
On macOS: brew install 64tass.
The target machine is selected by the BUILD_* flags at the top of
src/6502/mandelbr8.asm. Exactly one must be 1.
# C64 multicolor (default)
64tass -o build/c64.prg -a src/6502/mandelbr8.asm
# C64 contour (set BUILD_HIRES_CONTOUR = 1 in source first)
64tass -o build/c64-contour.prg -a src/6502/mandelbr8.asm
# C128, Plus/4, VIC-20, PET, CBM-II (toggle the matching BUILD_* flag)
64tass -o build/c128.prg -a src/6502/mandelbr8.asm
64tass -o build/ted.prg -a src/6502/mandelbr8.asm
64tass -o build/vic20.prg -a src/6502/mandelbr8.asm
64tass -o build/pet.prg -a src/6502/mandelbr8.asm
64tass -o build/cbm2.prg -a src/6502/mandelbr8.asm
# Atari XL/XE
64tass --output-exec=main --atari-xex \
-o build/atari.xex -a src/6502/mandelbr8.asm
# BBC Micro B (bootable SSD)
64tass -b -o build/beeb.ssd -a src/6502/mandelbr8.asmTunable build options at the top of src/6502/mandelbr8.asm:
BUILD_HIRES_CONTOUR, BUILD_MARIANI_SILVER_STRIPE, BUILD_LUMA_PALETTE,
BUILD_BENCHMARK, BUILD_PERIODICITY.
Each iteration:
zx, zy ← zx + cx, zy + cy
zx², zy² ← zx·zx, zy·zy
if zx² + zy² ≥ 4: diverged, stop
zx, zy ← zx² − zy², 2·zx·zy
Most 8-bit CPUs have no integer multiply, which makes naive implementations slow. This algorithm uses Q5.11 fixed-point arithmetic (5-bit signed integer part, 11-bit fraction; range [−16, +16)) with a fast 16×16 signed multiply. The trade-off is a limited zoom range: the precision floor is reached around 12× from the default view.
If the machine has at least 64 KB RAM, a 32 KB lookup table of squared
Q4.10 values is built at startup. This eliminates two of the three
multiplies per iteration, leaving only 2·zx·zy to compute via
smult12 (Toby Lobster's quarter-square multiply, ~234 cycles).
If a Kawari VIC-II replacement is detected on C64, hardware
multiplication via $D02F–$D033 registers is used, saving ~25 % of
total render time.
The Mandelbrot set fits inside a circle of radius 2, but iteration can produce intermediate magnitudes that exceed it. Q5.11 balances density against headroom for those intermediates.
Most targets render in two passes:
- Lo-res preview at e.g. 40×25 (one Mandelbrot point per cell). Cardioid and period-2-bulb fast tests skip the iteration loop for points known to be inside the set.
- Hi-res at full resolution. C64 uses stripe-wise Mariani-Silver border tracing; other targets render row by row.
The hi-res image is built in 40-pixel stripes (10 multicolor cells or 5 contour tiles per stripe). For each stripe:
- Border trace: compute the 92 perimeter samples (top + bottom rows plus 6 interior pixels of each side column).
- Recursive subdivision: pop a sub-rectangle from a stack. If its border samples are uniform, fill the interior with that iter without further computation. Otherwise split at the midpoint, compute the 6 pixels of the split column, and push the two halves.
- Sub-rectangles narrower than 6 pixels are brute-forced.
Default-view measurements on stock C64 (PAL):
- Multicolor: 102 s with stripe MS, 140 s without.
- Contour: 175 s with stripe MS, 261 s without.
VIC-II 4×8 multicolor blocks support 4 colors:
- 1 shared background (black, used for the in-set Mandelbrot region).
- 3 per-tile colors from the 16-color palette.
A histogram of each tile's iter values selects the 3 most common as that tile's per-tile colors. Some artifacts can remain depending on tile location in the complex plane.
BUILD_LUMA_PALETTE (default-on) reorders the palette so adjacent
iter values map to perceptually adjacent luminance. The same flag
also gives:
- C128 VDC: a luma-ordered 16-color palette mapped to the VDC CGA-style colors.
- TED (Plus/4 / C16): a uniform-luma palette. TED multicolor allows
only 2 cell-specific colors per 4×8 cell, so any luma variation
between cells creates visible stepping. Tunable via
TED_HI_LUMAin the source. - VIC-20: the 4 fixed multicolor slots reordered as black → blue → purple → white.
Creative Commons Attribution 4.0 International (CC BY 4.0).
https://creativecommons.org/licenses/by/4.0/
This project is based on / adapted from mandelbr8, originally created
by GitHub user 0x444454.
Original repository: https://github.com/0x444454/mandelbr8
The original repository appears to be no longer available on GitHub.
The original project was licensed under Creative Commons Attribution
(CC BY). This version contains modifications by Janne Tompuri /
jtompuri.