ULTIMATEEXPERIMENTSCommodore 64 · Work in progress

Existing emulator references

Development note: see current status, build instructions and installation for present qualification. Historical measurements below apply to their stated slice, not every current build.

Inspected 2026-10-14. The user clarified that the target is the original 8088 / IBM PC XT, implemented on the C64U's 6502/6510. References inform techniques; their capabilities never narrow the project's requirements. No code from the projects below has been incorporated into the runtime.

Closest host-CPU match: NES86

NES86 implements an 8086 on the NES in 6502 assembly, built with ca65/ld65. This is a real example of the direction we need: 6502 executing an x86 interpreter. It is not a C emulator of both processors, nor an emulator running on an 8088. Pin: d72debaa0a5daed533d994e087fa2963e1793820. Repository LICENSE is GPLv3; any source adaptation must retain notices and meet that license, rather than being treated as newly authored code.

Inspected modules provide these concrete references:

Module Useful technique Adaptation boundary
x86.s Separate fetch/decode/execute/write stages and explicit single-step entry Benchmark stage overhead; preserve our test ABI and bounded scheduler
decode.s Shared operand forms, opcode-indexed tables and split low/high dispatch tables Preserve original-8088 opcode profile and prefix semantics
reg.s Byte/word register maps and ModR/M register maps in zero page Account for our limited ZP ownership and existing register layout
alu.s, execute.s 8/16-bit arithmetic on an 8-bit host; separate flag computation Verify independently against hardware-generated 8088 cases, especially shifts, divide and decimal-adjust
mem.s Segmented-address interface separates operands from host memory mapping MMC5 maps cartridge RAM directly; REU requires DMA and coherent cached host pages. Replace mapping, do not port its mapper assumptions

The linker layout allocates 128 KiB guest RAM and cartridge ROM. The BIOS is tailored to ELKS and a serial console; it is not a DOS disk/CGA BIOS to drop in. Neither the project's advertised operation nor source inspection proves original-8088 conformance or our performance goal. Do not import reset defaults, segment-edge behavior, instruction timing, or undocumented-opcode behavior without tests. On the 6510, explicitly keep host decimal mode clear during binary arithmetic; NES code runs on a CPU whose decimal behavior differs.

Searches also returned Appler, PCjs and multi-CPU desktop emulators. Appler runs a 6502 emulator on an 8088, the opposite direction. The searches did not establish a separate, explicitly 8088/XT-focused 6502/6510 core. This is a search result, not proof that none exists.

Other useful references

Project and inspected pin License evidence Intended use and limits
8086tiny, c79ca2a34d96931d55ef724c815b289d0767ae3a MIT in license.txt and BIOS source header Compact C interpreter and NASM BIOS. Useful BIOS contracts, boot/INT 13h flow and decoder structure. C implementation is not our target runtime or a hardware oracle
XTulator, 211791d9b41871fb18079e37e696f6b95d2c3f88 CPU source header: GPLv2 or later Readable C flags/ModR/M/peripheral reference. Includes 80186 extensions; filter those out. GitHub's missing license metadata does not supersede the source header
emu2, 9d8698d06e67359bc8b230c2d19d5b1503e60819 GPLv2 LICENSE CPU/console reference; native DOS API emulation is a different architecture from our real DOS boot
MartyPC and SingleStepTests/8088 Pin and audit exact corpus/license when importing Prefer original-CPU evidence for correctness; desktop emulator performance does not predict C64U performance

8086tiny's BIOS source uses private 0F 00 through 0F 03 services and carries decoder lookup tables. Its BIOS/emulator coupling and memory layout differ from ours. If BIOS code is adapted later, record exact source ranges, modifications and MIT notices; do not adopt private instructions as generally executable guest opcodes. Our proposed native BIOS entry callbacks must still allow guest interrupt-vector hooks.

Historical 6502-hosted CP/M work

Dann McCreary's original 8080 Simulator for the 6502, KIM-1 Version and 1979 design article are author-provided scans. They illustrate compact opcode grouping and register simulation. Publication of scans does not establish a reusable open-source license; treat them as historical design reading.

Michael Steil's source-level examination of Roßmöller's C64 CP/M emulator traces it to McCreary's size-focused 8080 simulator. Its extra address translation and wrapper overhead are useful cautionary examples for a fast interpreter. It is not a full Z80 or x86 core, and its licensing history makes it unsuitable for unexamined copying. CP/M-65 is instead a native 6502 OS port, not execution of 8080 programs.

Implementation policy

  1. Implement the required original-8088 behavior, genuine DOS 3.3 boot and our PC device subset whether or not a reference provides it. The latest user clarification explicitly confirms this priority.
  2. Keep our CPU-state ABI, REU/cache/storage interfaces and Zig/Peddle behavior tests. Inspect useful routines before designing the next slice; do not make progress depend on finding reusable code.
  3. Use hardware-generated original-8088 cases and independently calculated defined flags as acceptance evidence. Comparing two emulators is supplementary evidence; they can share a bug.
  4. Before substantial code reuse, pin it, preserve its license/attribution, identify modifications and measure linked code/ZP/stack costs. No blanket project license choice is made by this research document.
  5. Keep downloads in project-local ignored scratch. The inspected NES86 file hashes are retained in the inspection record (local development record, not distributed); no upstream source is vendored by this research pass.