SID Detector v1.5.05

A Commodore 64 diagnostic utility that identifies 24+ variants of the SID (Sound Interface Device) chip — including real hardware, FPGA clones, microcontroller emulators, and PC emulators.

Original release: https://csdb.dk/release/?id=176909 Author: funfun/triangle 3532 Syntax: KickAssembler (converted from ACME original)

STORY.md — Full technical article: detection methods, reverse-engineered protocols, SIDFX secondary probing, self-modifying code patterns, hardware testing methodology, and lessons learned. For sceners, SID musicians, and hardware developers.

---

Screenshot

---

Hardware test rig

Representative photos of the chips and boards the detector is validated against — see pictures/ for the full set.

Reference chip collection: 6581 R2 / R3 / R4AR and 8580 R5 silicon on foam.

C64 with FM-YAM cartridge, Commodore FM Sound Module (SFX Sound Expander), and Ultimate II+ — used to verify the V1.4.x OPL $DF40/$DF50/$DF60 detection path.

---

What it does

When run on a C64 (or emulator), the program probes SID hardware registers, measures timing characteristics, and displays the results on screen. It detects:

Category	Chips / Variants
Real chips	6581 R2, R3, R4, R4AR · 8580
FPGA clones	FPGASID (6581 mode · 8580 mode)
Microcontroller	ARMSID · ARM2SID · Swinsid Ultimate · Swinsid Nano · Swinsid Micro · SIDKick-pico · KungFuSID · BackSID · PD SID · SIDFX · ULTISID (U64) · uSID64
Emulators	VICE 3.3 ResID 6581/8580 · VICE 3.3 FastSID · HOXS64 · Frodo · YACE64 · EMU64 · C64DBG
FM expansion	CBM SFX Sound Expander · FM-YAM (Yamaha OPL2 / YM3812 at $DF40)
MIDI interfaces	Sequential Circuits · Namesoft · DATEL/Siel/JMS · Passport · Maplin (6850 ACIA)
Machine type	C64 · C128 · TC64 (Turbo Chameleon 64)
Clock	PAL · NTSC
Stereo	Scans D400/D500/D600/D700/DE00–DFFF for additional SID chips
Info pages	Press I on the result screen for per-chip detail; CRSR LEFT/RIGHT to flip pages
Quality page	Press Q for a per-slot audio-quality fingerprint (sidcheck grade + $D418 decay) — one row per detected SID

Keys on the result screen

Key	Action
SPACE	Restart the full detection sequence (raster-stable; silences all voices)
I	Per-chip info pages (CRSR LEFT/RIGHT or B/M to flip; SPACE returns)
Q	Quality Fingerprint page (sidcheck grade + $D418 decay per slot)
D	Debug page — raw detection values, UCI/SIDFX state (D again → page 2)
R	README / help viewer (W/S to scroll)
T	SID sound test — plays the 3-voice melody on every detected slot
P	SID Tracker View — live per-voice display while music plays
L	TLR sid-detect2 second-SID detector (copied to $0801 and run)

---

Detection chain

The program runs a sequential detection pipeline at startup. Each step either identifies the chip and jumps to the result display, or falls through to the next step.

Step 1 — SIDFX (DETECTSIDFX)

Communicates with the SIDFX cartridge using its SCI serial protocol over the D41E/D41F register pins. Sends a "PNP" login packet ($80 $50 $4E $50) and reads back a 4-byte vendor/product ID. Checks for the expected signature $45 $4C $12 $58.

• data1 = $30 → SIDFX found
• data1 = $31 → SIDFX not present

When SIDFX is found, D41D (SW2/SW1/PLY) and D41E (SID1/SID2 types) are saved. After detection, sidfx_populate_sid_list populates sid_list: slot 1 = D400 with SID1 type, slot 2 = secondary address (D420/D500/DE00 per SW1 bits 5:4) with SID2 type. Before writing the secondary type, sfx_probe_skpico probes the secondary for SIDKick Pico identity: enters config mode by writing $FF to secondary+$1F, then reads VERSION_STR[0]='S' and [1]='K' at secondary+$1D — if confirmed, the slot type becomes $0B/$0E (SIDKick Pico 8580/6581) instead of the generic 6581/8580 from SIDFX D41E.

Step 2 — ARMSID / ARM2SID / Swinsid Ultimate (Checkarmsid)

Writes the ASCII string "DIS" to voice-3 registers (D41D/D41E/D41F), waits, then reads back D41B/D41C/D41D. Real SID chips do not echo writes; clone chips do:

Echo in D41B	Meaning
'S' = $53	Swinsid Ultimate
'N' = $4E	ARMSID family

For ARMSID, D41C and D41D further discriminate:

• D41C = 'O' ($4F) and D41D = 'R' ($53) → ARM2SID
• D41C = 'O' ($4F) and D41D ≠ 'R' → ARMSID

Self-modifying code: Checkarmsid patches the high-byte of all its SID register addresses at runtime (e.g. cas_d418+2, cas_d41D+2) so the same routine works for both D400 and D500 (stereo second-SID scan). This is the most complex routine in the codebase.

Steps 3a–3c — PD SID / BackSID / SIDKick-pico

These three checks run in sequence after ARMSID. All use D41D–D41F register probe protocols.

PD SID (checkpdsid): Writes 'P' to D41D, 'D' to D41E, reads D41E back — expects 'S'. data1 = $09.

BackSID (checkbacksid): Writes $42 to D41C, $B5 to D41D, $1D to D41E, reads D41F — expects $42 echoed. data1 = $0A.

SIDKick-pico (checkskpico): Enters config mode by writing $FF to D41F. The VERSION_STR pointer is manual — write $E0+i to D41E to select byte i, then read D41D. No auto-increment on reads.

D41E write	D41D read	Meaning
$E0	$53 = 'S'	byte[0] of VERSION_STR
$E1	$4B = 'K'	byte[1] of VERSION_STR

If both match → SIDKick-pico (data1 = $0B). Confirmed against firmware v0.22 DAC64. Full VERSION_STR: SK\x10\x09\x03\x0F0.22/DAC64.

Step 3d — FPGASID (checkfpgasid)

Enters FPGASID configuration mode by writing the magic cookie $81/$65 to D419/D41A, then sets bit 7 of D41E. Reads D419 and D41A back and checks for the identify signature $1D/$F5. If matched:

• D41F = $3F → FPGASID 8580 mode (data1 = $06)
• D41F = $00 → FPGASID 6581 mode (data1 = $07)

Step 4e — Real SID (checkrealsid, normal position)

Uses the voice-3 oscillator readback technique (reference: 1541 Ultimate detection code):

1. Writes $48 (gate bit set) to D412 (voice 3 control)
2. Shifts right and writes the sawtooth waveform value to D412
3. Reads D41B (oscillator 3 output) — real SIDs echo predictable values; emulators/empty sockets do not

Then reads D41B a second time and checks for $03. From that value, MODE6581/MODE8580 look-up tables classify the exact sub-revision.

Step 5 — Second SID scan (checksecondsid)

Uses the noise-waveform mirror test: activates noise waveform on voice 3, then reads D41B at $D400 + $20 offsets. A real SID at that address produces non-zero random values; a mirrored (non-SID) address always reads zero.

Scans: D400, D420, D440 … DE00, DE20, DEE0, DF00 … DFE0

Step 3e — uSID64 (checkusid64)

Runs after FPGASID, before the real SID check. Writes the config unlock sequence $F0 $10 $63 $00 $FF to D41F, then reads D41F twice with a ~3 ms gap.

• Both reads must be in $E0–$FC range (not $FF)
• Both reads must agree within $02 of each other (stable — the chip holds the value)

A decaying NOSID bus typically reads $FF (the last written value), or if it has drifted into $E0–$FE, the two reads will differ by more than $02. Either condition rejects the chip.

• data1 = $0D → uSID64

Step 0.25 — Real SID pre-check (checkrealsid, early)

Runs before SwinSID Nano to prevent false positives on real 6581/8580. Uses the voice-3 oscillator readback (see Step 4e). checkrealsid only writes to D412/D40F — never D41F — so it is safe to run early. If a real SID is confirmed here, checkswinsidnano is skipped entirely.

Step 5c — SwinSID Nano (checkswinsidnano)

Runs at Step 0.5 (before SIDFX), provided no real SID was found at Step 0.25. Uses D41B (OSC3) activity counting with noise waveform at maximum frequency ($FFFF):

Stage 1 — Change-count gate: Reads D41B 8 times back-to-back and counts how many consecutive pairs differ. A real 6581/8580 LFSR advances every CPU clock at $FFFF frequency — all 7 pairs always change (cnt = 7). The SwinSID Nano AVR updates at ~44 kHz (much slower than the 985 kHz C64 clock), so some reads catch the same LFSR value (cnt = 3–7, hits 7 in ~40% of windows). Stage 1 retries up to 3 times and only rejects if all attempts give cnt = 7 (guaranteed real-SID speed). P(all 3 fail for SwinSID Nano) ≈ 6%.

Stage 2 — Activity confirmation at 62 ms: After a 50 ms wait, counts changes in another 8-read window. Requires cnt ≥ 3. Filters out a fully-dead NOSID bus that happens to have passed Stage 1.

Known limitation: A C64 with an Ultimate II+ cartridge and virtual SID disabled generates FPGA-sourced bus noise at ~44 kHz that is indistinguishable from the SwinSID Nano oscillator. Such a setup will be reported as SwinSID Nano rather than NOSID.

• data1 = $08 → SwinSID Nano (or NOSID + U2+ with virtual SID off)

Step 5b — KungFuSID (checkkungfusid)

Runs after all other hardware checks have failed (last chance before SwinSID Nano / NOSID). Writes $A5 (firmware-update magic) to D41D, waits ~6 ms, reads D41D back.

Read back	Meaning
$5A	New firmware: ARM acknowledged the update-start magic → KungFuSID
$A5	Old firmware: register echoes the last write (all SID registers are stored in RAM array) → KungFuSID
anything else	Not KungFuSID

Old firmware reason: kff_read_handler returns SID[register_addr] for every address except the firmware-update register on new firmware. Writing $A5 stores it; reading back returns $A5. Real SID chips, ARMSID, FPGASID, and SIDKick all produce different values.

• data1 = $0C → KungFuSID

Step 6 — $D418 decay fingerprint (calcandloop + ArithmeticMean)

Sets the volume register D418 = $1F then counts CPU cycles until it decays to zero. Different emulators decay at different rates. The routine:

1. Samples 6 times (controlled by NumberInts = $06)
2. Averages the samples with ArithmeticMean (16-bit accumulator, integer division)
3. Compares the average against a MODE6581/MODE8580/MODEUNKN table to print the emulator name

This fingerprint distinguishes VICE ResID, VICE FastSID, HOXS64, Frodo, YACE64, and EMU64.

Step 7 — TLR baseline sweep (tlr_sweep, V1.5.01)

Runs only when no primary chip was identified (data4=$00). A family-agnostic scan adapted from TLR's sid-detect2: walks $D400–$D7E0 and $DE00–$DFE0 in $20 strides and, at each unclassified slot, runs a retriggered-sawtooth OSC3 count-up test (the readback must increment across three reads — proof an oscillator is advancing there). New finds are appended to sid_list as type $11 ("TLR-generic"); a dedupe pass collapses duplicates against the family-specific results, keeping the refined type.

Step 8 — FM expansion (checkfmyam, V1.4.x)

Probes $DF40/$DF50/$DF60 for a Yamaha OPL family chip (YM3526 / YM3812) — the CBM SFX Sound Expander and FM-YAM both decode there identically. Detection uses an (status & $E0) == 0 two-read check: a real OPL drives status into $00–$1F; open bus has the high bits set. Reported on the next free stereo row as DF40 SFX/FM FOUND. One detection path covers both products (see STORY.md §23).

Step 9 — MIDI cartridge (checkmidi, V1.4.45)

Probes the four documented C64 MIDI interfaces for a 6850 ACIA reset signature: write $03 (master reset) to the control register, then status reads & $73 == $02 (transmit-empty, nothing received, no errors). Address pairs, first-hit-wins: Sequential/Namesoft $DE00/$DE02, DATEL/Siel/JMS $DE04/$DE06, Passport $DE08, Maplin $DF00. A two-read consistency check rejects open-bus jitter; the $DF00 probe is guarded against SIDFX / U64 / SIDKick-pico FM / ARM2SID SFX- ownership. When no SID was found, the cart name overwrites the NOSID row. See STORY.md §24.

---

Machine type detection

PAL / NTSC (checkpalntsc)

Patches the NMI vector to RTI, then checks whether a raster IRQ fires at line $137 (which only exists on PAL machines with 312 lines). Result stored in KERNAL variable $02A6:

• 1 = PAL (~50 Hz)
• 0 = NTSC (~60 Hz)

All timing loops in the detection chain are calibrated to this value.

C64 / C128 / TC64 (check128)

1. Reads $D030 (C128 speed register; returns $FF on open C64 bus → C64 identified)
2. If not $FF, writes $2A to $D0FE and reads it back:
   • Returns $FC → Commodore 128
   • Returns other → Turbo Chameleon 64 (TC64)

Result stored in za7 ($A7).

---

Memory layout

Address	Contents
$0801	BASIC stub: SYS 9216 (→ $2400)
$1800	Embedded SID tune 1 (Triangle Intro, $1806 play)
$0A00	Embedded TLR sid-detect2 (copied to $0801 on L)
$2400	Main program — start: and all detection subroutines ($2400–~$5A99)
$5B00	tlr_sweep family-agnostic baseline scan (V1.5.01)
$6000	Detection result tables (num_sids, sid_list_l/h/t, sid_map), screen data, info-page text, string labels, colour table
$9200	SID Tracker View code (V1.4.33)
$A000	Embedded SID tune 2 (Delirious 9, under BASIC ROM)
$C000	Tracker shadow SID ($C000–$C01F)
$C020	Tune-selector segment (V1.4.35)
$C300	Quality Fingerprint page code + tables (V1.5.02)

Zero-page variables

Address	Name	Purpose
$A4	data1	Primary result (chip type code)
$A5	data2	Secondary result (echo char / sub-type)
$A6	data3	Tertiary result (ARM2SID 'R' discriminator)
$A7	za7	Machine type: $FF=C64, $FC=C128, other=TC64
$F7	sidnum_zp	Number of SID chips found
$F9–$FA	sptr_zp	SID base address pointer (e.g. $D4:$00)
$FC–$FD	mptr_zp	Mirror-scan address pointer

---

Result codes (data1)

Code	Chip
$01	6581
$02	8580
$04	Swinsid Ultimate
$05	ARMSID / ARM2SID
$06	FPGASID 8580
$07	FPGASID 6581
$08	Swinsid Nano
$09	PD SID
$0A	BackSID
$0B	SIDKick-pico (8580)
$0C	KungFuSID
$0D	uSID64
$0E	SIDKick-pico (6581)
$10	Second SID found
$20–$21/$24–$26	ULTISID 8580
$22–$23	ULTISID 6581
$30	SIDFX
$31	No SIDFX
$F0	No SID / Unknown

---

Screen layout (v1.5.05)

            siddetector v1.5.05

row  2: armsid.....:  [result]
row  3: swinsid....:  [result]
row  4: fpgasid....:  [result]
row  5: 6581 sid...:  [result]
row  6: 8580 sid...:  [result]
row  7: sidkick....:  [result]
row  8: backsid....:  [result]
row  9: kungfusid..:  [result]
row 10: pd sid.....:  [result]
row 11: nosid......:  [result]
row 12: sidfx......:  [result]
row 13: pal/ntsc...:  [PAL/NTSC]   [C64/C128/TC64]
row 14: usid64.....:  [result]
row 15: $d418 decay:  [value]
row 16: stereo sid.:  [address + chip name per slot]

Spacebar restarts the full detection sequence (raster-stable restart via $D012 spin). Pressing SPACE also silences all SID voices immediately.

Press P to launch the SID Tracker View — a dedicated screen that plays the built-in Triangle Intro (Michael Troelsen / Fun Fun 1988) and shows per-voice state live while music plays:

• NOTE / WAVE / GATE / ADSR / FREQ columns for voices 1–3
• VU bar per voice with a white → grey → yellow → orange → red gradient (voice 3 reads real $D41C ENV3; voices 1/2 use a software envelope follower with gate-edge detection)
• Live OSC3 scope across rows 15–22 (40-sample burst of $D41B)

Press P, SPACE, or Q on the tracker screen to stop and return to detection.

Technical note: SID voice registers are write-only, so to display what the player is writing to voices 1 and 2 the tracker patches the player binary at $1800–$1FFF, rewriting every STA $D4xx to STA $C0xx. The raster IRQ then copies $C000–$C01F → $D400–$D41F each frame after the play call, so audio is unaffected. On exit, an undo table restores the original bytes for a clean re-entry.

Press I to enter the info page for the detected chip. Navigate with CRSR LEFT/RIGHT; SPACE returns to the main screen. 17 pages are available (one per chip type), browsable in any order.

Press Q to enter the Quality Fingerprint page (added V1.5.02). It paints one row per detected SID, combining two orthogonal accuracy fingerprints:

D400 QUALITY 4/5 (GOOD  8580  ) D418=N15
D420 QUALITY 2/5 (BAD   ARMSID) D418=N18

• sidcheck grade (0–5, banded AWFUL / BAD / GOOD / BEST) — the combined-waveform OSC3 readback test from Wonderland XIII / Censor Designs, lifted from the VICE testprog suite and rebased to run per sid_list slot. The cycle-critical writes target absolute $D4xx operands that are runtime-patched per slot (a 36-site patch list generated at assemble time).
• $D418 decay (Nnn) — the same volume-decay measurement used on the main screen's row 15, surfaced for every slot so implementations can be compared side by side.

The chip-name column is resolved through sid_type_index, the single code→name table shared with the debug page. SPACE or Q returns to the main screen.

Note: the decay column reads N00 under VICE because reSID doesn't model $D418 read-decay like real silicon; the sidcheck grade is meaningful in both. Real-hardware decay values are captured by make hw_test (TEST 9). The sidcheck result table is PAL-calibrated.

---

Build

make          # assemble siddetector.asm → siddetector.prg  (requires Java + KickAss.jar)
make run      # launch in the patched WinVICE 3.9
make clean    # remove siddetector.prg

# Variant-specific launches (patched VICE only)
make run-armsid       make run-arm2sid      make run-swinu        make run-swinnano
make run-fpgasid8580  make run-fpgasid6581  make run-pdsid        make run-kungfusid
make run-backsid      make run-usid64       make run-sidfx        make run-skpico8580
make run-skpico6581   make run-none

# MixSID / stereo (8580 @ D400 + personality @ D420)
make stereo-armsid    make stereo-arm2sid   make stereo-swinu
make stereo-fpgasid   make stereo-sidfx

# Regression harnesses
make ci               # 32 unit tests       (~30 s)
make ci-full          # ci + 14 variant goldens  (~4 min)
make test-variants    # variant sweep alone
make update-variant-goldens    # after intentional UI or personality change

Requirements:

• Java runtime
• KickAssembler at C:/debugger/kickasm/KickAss.jar
• Patched WinVICE 3.9 at C:/Users/mit/claude/c64server/vice-sidvariant/GTK3VICE-3.9-win64/bin/x64sc.exe (build recipe: docs/VICE_PROXY_BUILD.md; patch: patches/vice-sidvariant-v1.patch). All VICE-based tests and make run-* / stereo-* targets require this binary — the stock VICE doesn't know the -sidvariant flag.

---

Chip-variant testing without hardware

The repo ships a patched fork of VICE 3.9 (in ../vice-sidvariant/) with a -sidvariant <name> CLI flag. Each emulated SID slot can wear a chip-family personality that responds to that chip's detection magic-cookie protocol — ARMSID, ARM2SID, SwinSID U/Nano, FPGASID 6581/8580, PDsid, KungFuSID, BackSID, SIDKick-pico 6581/8580, SIDFX, uSID64. Audio still comes from ResID; only the detection protocol is intercepted.

This makes CI exercise every chip family without plugging anything into a real C64. The window title shows the active personality (e.g. VICE (C64SC) [SidVariant=sidfx]) so you always know what's loaded.

make ci-full runs 32 unit tests + a 14-case variant sweep that byte-diffs each variant's detection screen against a stored golden (in tests/variant_goldens/). If a regression changes what a variant looks like, the diff is printed and CI fails loudly.

Full design rationale and per-chip protocol tables:

• docs/ARMSID_PROXY_PLAN.md — ultraplan with the 9-phase build-out.
• docs/VICE_PROXY_BUILD.md — reproducible build from MSYS2.
• docs/VICE_PROXY_USAGE.md — flag catalogue + make targets.
• docs/test_matrix.html — combined HW + VICE test dashboard (open in browser).
• patches/vice-sidvariant-v1.patch — ~2 kLOC source diff against pristine VICE 3.9 (upstream-ready, GPLv2+ inherited).

---

Unit testing

Tests are written in 6502 assembly (KickAssembler syntax) and run inside WinVICE. Each test preset the relevant zero-page inputs (data1, data2, data3, za7), calls an embedded copy of the dispatch logic, compares the returned result code against the expected value, and displays PASS / FAIL on the C64 screen. The final pass count is written to $07E8 (off-screen RAM) for inspection in the VICE monitor.

Running tests

make test           # ArithmeticMean unit tests       (4 cases)
make test_dispatch  # ARMSID / FPGASID dispatch tests (8 cases)
make test_suite     # Full suite — all scenarios      (43 cases)

After VICE opens, all results are visible on screen immediately. To check the pass count in the VICE monitor (Alt+M):

mem $07E8 $07E8    # shows pass count; $2B (43) = all passed for test_suite

Test files

File	Tests	Covers
tests/test_arith.asm	4	ArithmeticMean — pure computation
tests/test_dispatch.asm	8	ARMSID / ARM2SID / FPGASID dispatch
tests/test_suite.asm	43	All dispatch stages (see table below)

Each file has a matching .mon VICE moncommands file that loads symbols and sets a breakpoint at td_spin (the completion spin-loop).

Full suite coverage (test_suite.asm)

Section	Tests	Stage	Inputs → Expected result
S1	T01–T03	Machine type	za7=$FF → C64 · $FC → C128 · other → TC64
S2	T04–T05	SIDFX	data1=$30 → found · $31 → not found
S3	T06–T10	Swinsid/ARMSID	$04 → Swinsid-U · $05/$4F/$53 → ARM2SID · $05/$4F/other → ARMSID · no-match cases
S4	T11–T13	FPGASID	$06 → 8580 · $07 → 6581 · other → no match
S5	T14–T16	Real SID	$01 → 6581 · $02 → 8580 · other → no match
S6	T17–T18	Second SID / no sound	$10 → second SID · other → no sound
S7	T19–T22	ArithmeticMean	[10,20,30]=20 · [5×6]=5 · [100,50,75,25]=62 · empty=0
S8	T23	FPGA stereo	data1=$06 at $D500 → recorded in sid_list
S9	T24–T27	New chips	$09 → PDsid · $0A → BackSID · $0B → SIDKick-pico · $0C → KungFuSID
S10	T28–T29	ARM2SID SFX	emul_mode=$01 → SFX only · $02 → SID+SFX
S11	T30–T31	SKpico FM	skpico_fm=$04/$05 → FM Sound Expander at $DF00
S12	T32	FM-YAM OPL2	fmyam_detected=$01 → FM-YAM/OPL2 at $DF40
S13	T33–T35	Quality band	score 0 → AWFUL · 5 → BEST · $FF → BAD-clamp
S14	T36–T43	Chip-type index	sid_type_index code→slot: $01→6581 · $02→8580 · $08→Nano · $09→PDsid · $0E→SKpico-6581 · $30→SIDFX · $F0→NoSID · $0F→UNKWN

Design approach

Because the actual detection routines (Checkarmsid, checkfpgasid, etc.) probe real SID hardware registers, they cannot be tested deterministically inside VICE — VICE's SID emulates an 8580 and returns fixed values. The tests therefore target the dispatch logic: the code that reads data1/data2/data3 (already set by the detection routines) and branches to the correct chip identification. Each dispatch routine is an embedded copy of the relevant branch block from siddetector.asm, with jsr $AB1E / jsr $E50C / jmp end replaced by a result-code write to dispatch_result.

Adding a test

1. Identify the zero-page inputs for the scenario
2. Add a dispatch routine that mirrors the branch logic from siddetector.asm
3. Write a test block: set inputs → call dispatch → compare dispatch_result → print PASS/FAIL
4. Increment pass_count on pass
5. Update the total expected count in test_done (cmp #N)

What remains untested

Area	Why
Checkarmsid hardware probe	Writes to SID regs; VICE returns fixed 8580 values
checkfpgasid magic-cookie	Config mode only exists on real FPGASID hardware
checkrealsid OSC3 readback	Depends on real sawtooth waveform decay
checksecondsid noise mirror	Depends on real noise-waveform $D41B randomness
calcandloop decay timing	Emulator timing differs from hardware by design

---

Known issues / TODO

• Added V1.5.05: Single source of truth for chip-type → name. The Q page and the debug page previously carried two independent code→name mappings that had drifted (that drift produced the V1.5.04 $01/$02 swap and $0E mismap). They now share one sid_type_index resolver + a 17-byte sid_code_to_slot table feeding row-aligned sidname_short_* (6-char Q page) and sidname_long_* (debug page) tables — adding a chip is one row in each and they cannot disagree. ULTISID ($20–$26) stays special-cased in each printer (debug names all 7 filter-curve variants; the 6-char Q page only distinguishes 8580 vs 6581 family). Unit suite grown to 43 tests (T36–T43 exercise sid_type_index directly — the keystone guard against this drift class); hw_test.py TEST 9 captures the per-slot Q-page sidcheck grade + $D418 decay off real hardware into the run report. Binary shrank 49222 → 49184 bytes.
• Fixed V1.5.04: Q-page chip-name table holes + a pre-existing debug-page swap. quality_print_chiptype routed five valid type codes to "UNKWN" — $09 PDsid, $0E SIDKick-pico 6581 (was mislabelled "USID64"), $10 secondsid, and $20/$21/$22 ULTISID — now PDSID/SKPI65/2NDSID/ULTI85/ULTI65. Separately, dbg_print_sid_typename had type $01 and $02 mapped to the wrong strings (6581 shown as "8580 FOUND" and vice-versa) since the routine was written; the canonical convention is $01=6581, $02=8580 (line 753 / 5894). Also retired do_quit/goodbye_text/EXITINTRO (~30 bytes dead code since V1.5.02 rebound the Q key), and fixed the taskkill //F quoting in the new smoke scripts.
• Fixed V1.5.03: Two latent Q-page bugs surfaced by interactive WinVICE smoke-testing (neither was touched by the unit suite or the variant goldens). (1) qc_pt_ptr was declared in the $C300 code segment, not zero page — KickAssembler silently truncates a non-ZP label in (label),y indirect-indexed addressing to its low byte, so sta (qc_pt_ptr),y assembled as sta ($A9),y and corrupted memory on every Q-press; moved to $C1. (2) calcandloop_q inherited the txs/tsx "save X" trick from calcandloop, which is only safe because the original tail-calls jmp funny_print and never returns — the Q-page copy returns via rts, so each iteration buried the return address; fixed by restoring X from a zero-page slot instead.
• Added V1.5.02: Quality Fingerprint page (Q key). One row per detected SID combining two orthogonal accuracy fingerprints: a sidcheck combined-waveform OSC3 grade (0–5, banded AWFUL/BAD/GOOD/BEST, lifted from the Wonderland XIII / Censor Designs VICE testprog) and the $D418 volume decay. The cycle-critical sidcheck writes target absolute $D4xx operands that are runtime-patched per sid_list slot via a 36-site patch list generated at assemble time by a qrec() macro, so the original's timing is preserved across all eight possible slots. Lives in a fresh $C300 RAM segment. SPACE or Q returns to the main screen.
• Added V1.5.01: TLR family-agnostic baseline sweep (tlr_sweep, adapted from TLR's sid-detect2). Walks $D400–$D7E0 and $DE00–$DFE0 in $20 strides and, at every unclassified slot, runs a retriggered-sawtooth count-up test (OSC3 must increment across three reads) before the family-specific scans. New finds are appended to sid_list as type $11 ("TLR-generic"); a dedupe pass collapses duplicates, keeping the refined type. Gated to data4=$00 (no primary chip identified) — a broader gate false-positives on mirrors and disturbs ARMSID DIS state.
• Added V1.4.45: MIDI cartridge detection. New checkmidi routine probes the four documented C64 MIDI carts (codebase.c64.org reference) via the 6850 ACIA master-reset signature: Sequential Circuits / Namesoft at $DE00/$DE02, DATEL/Siel/JMS at $DE04/$DE06, Passport at $DE08, Maplin at $DF00. First-hit wins (only one MIDI cart can be attached at a time per the codebase reference). Result is printed on row 11 (the NOSID line) at column 13, aligned with the other detection rows. Sequential and Namesoft share the polled-read fingerprint (only IRQ vs NMI line routing differs) and are both reported as SEQUENTIAL MIDI. The $DF00 (Maplin) probe is guarded against SIDFX / U64 / SIDKick-pico FM / ARM2SID SFX- ownership. The patched WinVICE 3.9 was rebuilt with --enable-midi to support headless make run-midi-{sequential,passport,datel,namesoft,maplin} and 5 new variant_smoke.py golden-fingerprinted test cases. Side-fix: the legacy diagnostic write at $5801 (which was clobbering bytes overlapping the MIDI string area) is now stored at a labelled diag_step4_result byte. variant_smoke.py also gained a -default flag to insulate runs from vice.ini settings persisted by interactive make stereo-* / make sfx sessions.
• Added V1.4.44: Polish + test infra. Restart progress bar shares row 24 with the "*** RESTARTING ***" banner instead of using a separate row (bar fills outside the banner range, recolours banner letters green inside it — one row total). Behavioural U64 fallback: is_u64 = 1 when the fingerprint scan finds 4+ slots even if the $DF1F UCI probe at start: reads $FF (UCI disabled / remapped). "TUNEFUL EIGHT READY ON U64" banner overlays row 11 (NOSID...:) when sidnum_zp == 8 AND is_u64 == 1. $01 = 6581 / $02 = 8580 is now canonical across code + docs; six stale comment sites and two reversed branches in u64_fingerprint_scan / uci_type_for_addr corrected (your ULTISID slots now correctly display as 6581 INT). utfa_drain capped at 256 iterations (was unbounded — could hang when UCI was unresponsive). New scripts/check_memorymap.py [--fix] verifies addresses in docs/MEMORYMAP.md against the live siddetector.sym; wired into make ci as a doc-drift guard. New make test-tuneful-eight runs a non-destructive end-to-end U64 8-SID hardware test: snapshot the live config, apply bin/tt8-ultimate.cfg, run siddetector, verify sidnum_zp == 8 + is_u64 = 1 (all 8 slots labelled 6581 INT), restore the pre-test config in a finally block.
• Added V1.4.37: Write-coupling scan (u64_fingerprint_scan) for 8-SID Ultimate 64 "Tuneful Eight" configurations — independently verified to find all 8 slots (D400 D420 D480 D4A0 D500 D520 D580 D5A0) on real hardware across three consecutive runs. The existing checksecondsid noise-mirror trick falsely rejects U64 UltiSID slots because every FPGA slot runs its voice 3 oscillator continuously, so reads of candidate $D41B are non-zero even when primary noise is enabled. The new scan instead silences primary D400 voice 3, snapshots each already-found slot's OSC3 baseline per candidate, writes a saw+gate activation to the candidate, and rejects only if the activation propagates to primary or any found slot's OSC3 (i.e. the candidate is a write-mirror). is_u64 ($DF1F != $FF) gate dropped because U64 configs that disable UCI return $FF there yet still have multi-SID hardware; on a single-SID C64 every candidate write mirrors to primary, so the test correctly rejects all candidates anyway. Bypassed uci_type_for_addr's UCI-FIFO drain loop (hangs forever when $DF1C/$DF1E both return $FF) — chip-type comes from a direct checkrealsid call.
• Added V1.4.36: 8-SID enumeration for Ultimate 64 "Tuneful Eight" config. num_sids / sid_list_l/h/t extended from 8 to 9 bytes (slot 0 reserved, slots 1..8 active); s_s_add cap raised from 7 to 8; csfp_l_l_found (FPGASID-stereo path) cap raised from 7 to 8; sidstereo_print now renders up to 8 SID rows at screen rows 16..23. Row 24 stays as the action bar. U64 detection itself was already in place via $DF1F != $FF → is_u64 = 1. Verified the hardware exposes 8 addressable SIDs by cross-checking with TLR's sid-detect2.
• Added V1.4.35: Second tune in tracker view (1/2 keys) and a progress bar across the SPACE-restart wait. Triangle Intro stays at $1800; Delirious 9 (Troelsen / Genesis Project, 1990) is SIDwinder-relocated to $A000–$B39B and embedded as a separate segment. tune_select patches the IRQ play / init JSR operands, the tracker_patch_once scan range, and the title row pointer; pressing 1 / 2 in the tracker view stops IRQ play, silences SID, repatches the new player and re-inits in place. BASIC ROM is banked out ($01=$36) only inside the IRQ play call and around tune_player_init / tracker_patch_once so $AB1E (BASIC STROUT) stays usable elsewhere. The new tune-management block lives in a fresh $C020 segment above the shadow SID. The 2 s SPACE-restart dwell now fills row 23 with green solid blocks (one cell per ~54 ms) instead of busy-waiting silently.
• Added V1.4.33: SID Tracker View (P key) — dedicated screen during music playback showing per-voice NOTE/WAVE/GATE/ADSR/FREQ, VU bars with white→red gradient, and live OSC3 scope for voice 3. Player writes are redirected to shadow RAM at $C000–$C01F via one-shot binary patch so voice 1/2 registers can be read back (normally write-only). Tracker code lives at $9200 (below BASIC ROM) to stay CPU-visible without bank switching. See MEMORYMAP.md.
• Fixed V1.3.84: SIDFX secondary D420 probing: (1) removed D41D echo test — SIDFX write-buffers unmapped registers ($1D–$1F), causing any chip at D420 to echo back the written value (not chip-specific). (2) When primary SID is ARMSID, skip DIS probe at D420 entirely — ARMSID snoops CS2 DIS writes and drives $4E aggressively on all D4xx bus reads, contaminating D43B. Falls back to SIDFX-reported type (6581/8580). DIS probe still works for D420 with non-ARMSID primary, and for D5xx+ regardless of primary. Added D41B ACK in sfx_probe_dis_echo before secondary reads (harmless for non-ARMSID primaries).
• Fixed V1.3.83: Detection confidence indicator — if checkrealsid needed retries due to VIC bad-line DMA steals, a * is appended after "6581 FOUND"/"8580 FOUND" on the main screen. retry_zp ($B2) tracks how many of the 3 attempts were used.
• Fixed V1.3.81: Multi-SID sound test now plays the full 3-voice melody on every detected SID slot (not just a triangle tone). snd_patch_page self-modifies all 31 sta $D4xx instructions in st_soundtest to the target SID page.
• Fixed V1.3.80: Stereo ARMSID@D400 + SwinSID U/ARMSID@D5xx — s_s_arm_call_real now allows sfx_probe_dis_echo when primary is ARMSID (data4=$05); the probe reads from candidate+$1B so D400 ARMSID snooping the DIS writes does not corrupt the result. Requires dual ARMSID/SwinSID U hardware to verify.
• Fixed V1.3.79: SwinSID Ultimate fiktivloop false positive — AVR OSC3 returns 0 with noise enabled, causing checksecondsid to falsely detect D500 as a second SID. Fixed by skipping fiktivloop when primary is SwinSID U (data4=$04).
• Fixed V1.3.79: Stereo 6581@D400 + SwinSID U@D500 — s_s_arm_call_real now tries sfx_probe_dis_echo before checkrealsid when primary is a real SID; SwinSID U echo returns 'S' and is correctly identified.
• Fixed V1.3.73: 8580@D400 + ARMSID@D420 (MixSID, C09 config) now correctly detected
• Fixed V1.3.74: 6581@D400 + ARMSID@D420 (MixSID, C08 config) now correctly detected
• ARM2SID stereo D400+D500 not yet verified on hardware
• Stereo slots D700 and DF00 not yet tested on hardware
• Swinsid Nano with NOSID+U2+ (Ultimate II+ with virtual SID off) is indistinguishable — reported as SwinSID Nano; accepted limitation
• FPGASID in SIDFX SID1 slot is undetectable: SIDFX drives D419/D41A (POT registers) with real joystick data, masking FPGASID's identify-mode readback signature; additionally, the SIDFX SCI state machine reacts to D41E writes, disrupting the magic-cookie handshake. No software workaround is possible. Reported as SIDFX with unknown SID type.
• Quality page $D418 decay reads N00 under VICE — reSID emulates the audio path faithfully but does not model the analog volume-register read-decay the measurement depends on. The sidcheck grade is meaningful under emulation; real decay values come from make hw_test on a U64. Not yet captured/calibrated on hardware.
• Quality page sidcheck result table is PAL-calibrated; NTSC scores may shift. Per-family expected-grade reference not yet built (awaiting hardware decay data).